Appl Health Econ Health Policy

https://doi.org/10.1007/s40258-018-0398-2

ORIGINAL RESEARCH ARTICLE

Impact of Reducing Glycated Hemoglobin on Healthcare Costs

Among a Population with Uncontrolled Diabetes
Megha Bansal1 • Mona Shah1 • Brian Reilly2 • Susan Willman3 • Max Gill1 • Francine R. Kaufman1

Ó Springer International Publishing AG, part of Springer Nature 2018

Abstract Results Of the 3,197 patients who had a first A1C C 9%,
Introduction Glycated hemoglobin (A1C) is considered a 2,273 patients (71%) had a decrease in A1C (Decreasers)
‘‘gold standard’’ measure of glycemic control in patients and 924 patients (27%) had an increase in A1C (Non-de-
with diabetes and is correlated with a lower risk of diabetes creasers). After matching, we compared 912 Decreasers to
complications and cost savings. This retrospective claims- 912 Non-decreasers. Patients in the former group had
analysis assessed the impact of A1C reduction on health- average annual healthcare costs that were 24% lower dur-
care costs in patients with uncontrolled Type 1 and Type 2 ing the first year of follow-up and 17% lower during the
diabetes. second year of follow-up, compared to patients whose A1C
Methods Using a large repository of US health plan did not decrease. This reflected a savings of US$2503 and
administrative data linked to A1C values, patients with a US$1690, respectively. For both time periods, the outpa-
diabetes diagnosis and at least two A1C values between 1 tient category was the largest contributor to cost savings.
January 2009 and 31 December 2014 were selected to Discussion In our analysis, A1C reduction among patients
identify changes in A1C and associated changes in with T1DM and T2DM was associated with slower growth
healthcare expenditure. We used all medical and pharmacy in healthcare costs within 1–2 years. These findings suggest
claims to calculate direct healthcare costs from 1 year prior that programs aimed at reducing A1C over a short time-
to the index A1C to 2 years after the index A1C. A frame may lead to substantial savings and may be worth
propensity score method was used to match patients with pursuing by health plans and other payers.
decreased A1C to patients whose A1C did not decrease,
based on potentially confounding variables. Then, a gen-
eralized linear model regression was used to estimate the
Key Points for Decision Makers
difference-in-difference (DD) effect on costs between the
two groups.
In a large cohort of patients with uncontrolled
diabetes, decreases in A1C were associated with
healthcare cost savings within a 2-year period.
Electronic supplementary material The online version of this
article (https://doi.org/10.1007/s40258-018-0398-2) contains supple- Diabetes management programs that focus on
mentary material, which is available to authorized users. lowering A1C may control costs within 1–2 years.

& Megha Bansal

megha.bansal@medtronic.com
1
Medtronic, 18000 Devonshire Street, Northridge, CA 91325,
USA
2
Medtronic, 18302 Talavera Ridge, San Antonio, TX 78257,
USA
3
Medtronic, 3033 Campus Drive, Plymouth, MN 55441, USA
 M. Bansal et al.

1 Introduction Although many studies have evaluated the impact of

glycemic control on long-term cost savings (e.g., over a
Diabetes, a disease marked by high levels of blood glucose lifetime) [5, 10, 11], value-based healthcare programs are
resulting from deficiencies in insulin production and/or also keenly interested in identifying the costs associated
action, is the seventh leading cause of death in the USA [1]. with short-term outcomes (e.g., 1- to 3-year) in patients
A major public health concern and economic burden both with diabetes. Several studies have found that better gly-
globally and nationally, diabetes accounted for $245 billion cemic control or A1C among these patients may be asso-
in direct and indirect costs (costs are in US $) in the USA in ciated with lower healthcare resource use and cost in the
2012 [2]. In 2015, the prevalence estimate of diagnosed short-term [12–17]. However, limitations of these studies
and undiagnosed diabetes in the USA was 30.2 million include relatively small sample size [15], limited geo-
adults (18 years of age or older), or 12.2% of the popula- graphic focus [16, 18], lack of control for disease severity
tion [1]. The prevalence is expected to more than double and other variables [14], a focus only on inpatient costs
over the next three decades, with approximately 33% of the [15, 19], and data sets that are more than two decades old
population projected to have diabetes by 2050 [3]. [12, 19].
Diabetes treatment, such as administration of oral To supplement the findings of earlier studies and address
medication or insulin therapy, regular glucose monitoring, some of the limitations described above, a population-
and dietary changes can have a significant impact on a based, retrospective, longitudinal claims analysis of
person’s daily routine [4]. Early in the disease progression, patients with uncontrolled diabetes (A1C C 9%) was con-
increasing insulin sensitivity through diet, exercise, and ducted to compare total healthcare costs over a 3-year time
weight management is often effective; however, when period to each patient’s changes in A1C in the first year.
these measures fail, use of antidiabetic agents can often This study was designed to test the hypothesis that, among
help patients achieve glycemic goals. Healthy lifestyle and patients with diabetes, a reduction in A1C is correlated
proper medical interventions can prevent high medical with slower growth in total healthcare costs over the short
costs and improve patients’ quality of life [4]. term.
The goal of treatment is to maintain blood glucose at
normal levels to reduce the risk and number of diabetes-
related complications, many of which may have lasting, 2 Methods
serious health consequences. For example, diabetes is a
leading cause of blindness, end-stage renal disease, and 2.1 Data Source
non-traumatic lower limb amputation, as well as a major
risk factor for coronary artery disease and stroke [1]. For this analysis, we utilized the de-identified Optum
Numerous research studies have shown that aggressive ClinformaticsÒ Data Mart database (OptumInsight, Eden
glycemic control can reduce these long-term complications Prairie, MN, USA) for a subset of commercially insured
in patients with diabetes and help control the costs asso- and Medicare Advantage plan members with diabetes,
ciated with their care [5, 6]. based on either diagnosis codes or use of antidiabetic
Because A1C, a clinical measure of ambient blood medications, including insulin. More than 6 million mem-
glucose concentrations over the previous 3 months, is bers, comprised of both subscribers and their dependents,
recognized as a reliable measure of chronic hyperglycemia were part of the initial data source. The database contains
and correlates well with the risk of diabetes complications, de-identified, person-specific data including insurance
quarterly A1C measurement is the standard of care for claims submitted by all providers of medical care or
testing and monitoring diabetes [7]. American Diabetes treatments [including pharmacies and durable medical
Association (ADA) guidelines recommend reducing A1C to equipment (DME) suppliers]. It also contains A1C values
\7% to delay the onset of diabetes-related complications, collected from select laboratory providers. Institutional
and the National Committee for Quality Assurance review board oversight was not necessary because the
(NCQA) has established a threshold A1C value of [9.0% database was de-identified in compliance with HIPAA
to identify patients with poor glycemic control [7, 8]. rules.
Unfortunately, relatively recent data indicate that 7–14% of
patients with diabetes do not undergo regular A1C testing, 2.2 Study Design
and, when tested, approximately 34–44% of commercially
insured patients, 27% of patients with Medicare, and 45% A population-based, longitudinal analysis that compared
of patients with Medicaid have A1C C 9% [9]. healthcare costs over a 3-year timeline to A1C change
during a baseline period (1 year prior to the index date) was
conducted. (Fig. 1) Patients were divided into two groups:
Impact of Reducing Glycated Hemoglobin on Healthcare Costs

2.4 Measures

2.4.1 Costs

Total direct healthcare costs were computed by summing

the costs associated with all medical and pharmacy claims.
Total direct costs included inpatient, outpatient (e.g., office
visits, outpatient surgery), emergency room (ER) services,
pharmacy (drugs and medical supplies), insulin pump costs
(not shown) and blood glucose meter costs (not shown).
Fig. 1 Study design. CY calendar year, A1C glycated hemoglobin Medical costs consisted of inpatient, outpatient, and ER
services. To account for differences in pricing across health
those whose A1C decreased by any amount during the plans and provider contracts, the database applies standard
baseline period (‘‘Decreasers’’) and those whose A1C pricing algorithms (adjusted to calendar year 2014) to all
increased/did not change (‘‘Non-decreasers’’). claims, so that claims can be considered to have been paid
by a single payer with a consistent fee schedule. According
2.3 Cohort Selection to Optum, the standard price is an estimate of the allowed
amount (the insurance paid amount plus the amount the
Using the database described above, we selected all patient is responsible for), since the amount payers actually
patients who, between 1 January 2009 and 31 December pay for a service varies widely based on contractual
2014, had at least two A1C measurements within 180–240 agreements with providers, geographic differences, etc.
days of each other, with the index date set as the data [21].
associated with the second A1C. In order to study those Outpatient costs were further broken down by type of
with uncontrolled diabetes, the study sample was narrowed service (TOS) codes listed in medical claims. TOS codes
to include only those with an initial A1C C 9%. Patients from medical tables were used to determine the type of
also had to be continuously enrolled with medical and outpatient service(s) each patient received. There were
pharmacy benefits through a commercial health insurance codes related to ancillary services, services provided by an
plan or Medicare Advantage health plan for 1 year prior outpatient facility, and professional services. Categories
through 2 years after the index date. We defined continu- were created by combining related services. For example,
ous health plan enrollment as a gap in enrollment that was all services related to radiology, laboratory, and pathology
less than 30 days in each half of the year. Patients were were combined under a diagnostics category. Other cate-
also required to be C 18 years of age 1 year prior to the gories created included ancillary services related to DME,
index date. office visits, outpatient emergency room, outpatient sur-
We further narrowed the sample to include only those gery, outpatient—other, and other services (i.e., rehabili-
who, within 1 year prior to the index date, had (1) a min- tation, mental health, and long-term care).
imum of one inpatient admission or two outpatient visits
with a diabetes diagnosis (International Classification of 2.4.2 Demographic and Insurance Plan Data
Diseases, Ninth Revision Clinical Modification [ICD-9-
CM] 250.XX) or (2) a minimum of one prescription for Demographic data collected during the baseline period
antidiabetic medication/insulin. included age, sex, and geographic region within USA. Age
Exclusion criteria included patients with cancer, HIV, was capped at 90 years of age to protect patient identities.
metastatic tumor, a transplant, major complications, nega- Measures also included whether the individual was enrol-
tive total costs, missing outpatient costs, and/or missing led in a commercial insurance plan or Medicare Advantage
information regarding sex, as well as those whose diabetes plan.
was not classified as type 1 diabetes mellitus (T1DM) or
type 2 diabetes mellitus (T2DM), who had a secondary 2.4.3 A1C
payer (coordination of benefits) anytime during the study
period, and/or who were missing components required to For a subset of all laboratory tests, actual A1C laboratory
calculate the Charlson Comorbidity Index (CCI) score [20]. results were available in the database. Each patient’s first
and second A1C measures within 180–240 days were
chosen for analysis.
 M. Bansal et al.

2.4.4 Health Status categorization was selected based on the clinical signifi-
cance of at least a 1 percentage point decrease.
Health status was defined in several ways. Diabetes Then, we conducted a test of the triple difference, by
severity was determined by presence of macrovascular and including a triple interaction term along with all of the
microvascular complications during the baseline period, as bivariate interactions, where a negative coefficient indi-
well as the Diabetes Complication Severity Index (DCSI) cates that those with at least a 1 percentage point decrease
[20, 22]. Overall health status was determined by the CCI in A1C had slower growth in costs compared to those with
score. \1 percentage point decrease in A1C (Online Supple-
mentary Material, Appendix 3)
2.5 Statistical Analyses Analyses were performed using SAS v.9.3 (Cary, NC,
USA) software.
To control for confounding factors, a propensity score
method was used to match patients with a decreased A1C
to patients whose A1c did not decrease. Propensity scores 3 Results
were estimated based on logistic regression with baseline
predictors of age, sex, insurance, presence of macrovas- 3.1 Cohort Selection and Study Groups
cular or microvascular complications, geographic region,
presence of inpatient costs, study start year, CCI score, Of the 50,021 patients having two A1C values and meeting
baseline total costs, and endocrinologist or primary care the inclusion criteria, 33,909 had a diabetes diagnosis. Of
physician (PCP) usage. Then, nearest neighbor matching these, 2,773 were excluded based on the criteria described
without replacement was done to select equal-sized study in the Methods section. This left 31,136 patients, of whom
groups of Decreasers and Non-decreasers, based on their 10.3% or 3,197 patients had a first A1C C 9% (Fig. 2).
propensity score falling within a range/caliper of 0.20 of During the baseline period (i.e., 1 year prior to the second
the pooled standard deviation of logit of the propensity A1C measurement), 2,273 patients (71%) in the study
score (Online Supplementary Material, Appendix 1). sample were categorized as Decreasers and 924 patients
A generalized linear model regression was used to (29%) were categorized as Non-decreasers.
estimate the difference-in-difference (DD) effect on costs,
which allowed comparison of the change in costs between 3.2 Baseline Characteristics
the Decreaser and Non-decreaser groups from the baseline
period to the follow-up period. Using time and group The very few differences observed between the study
(‘‘Decreaser’’ or ‘‘Non-decreaser’’) as covariates, the groups were seen mainly within demographic characteris-
interaction term of time and group represents the DD tics. Decreasers were less likely to be female (41 vs. 46%)
estimate of the impact of decreasing A1C on change in
healthcare costs. Costs were assumed to be gamma dis-
tributed with a log link function. These assumptions led to
regression coefficients that are logarithmically scaled. The
direction and magnitude of the regression coefficients
provide an indication of the independent variable effects.
Positive or negative coefficient values indicate whether a
variable is associated with increasing or decreasing costs,
respectively. The exponential of coefficient values is an
approximation of the cost multiplier for that variable
(Online Supplementary Material, Appendix 2).
In the basic model, the DD is assumed to be identical
across all the groups of Decreasers and Non-decreasers,
and the regression merely reports the average effect.
However, in reality, the DD effect may be a function of the
magnitude of A1C decrease. To test this, we repeated the
propensity score matching for two sub-groups of Decrea-
sers compared to Non-decreasers: (1) those who decreased
A1C by 1 or more percentage points, and (2) those who
Fig. 2 Selection of the study sample. *One patient with [$5M total
decreased A1C by less than 1 percentage point. This healthcare costs during the first year of follow-up was excluded. A1C
glycated hemoglobin
Impact of Reducing Glycated Hemoglobin on Healthcare Costs

and were older, with an average age of 56.3 years among When a propensity score method was used to match the
Decreasers and 54.8 years among Non-decreasers. Disease Decreaser and Non-decreaser groups with respect to
severity was very similar in the two groups, in terms of baseline characteristics, there were 912 patients in each
similar proportions of macrovascular or microvascular group with similar characteristics, as shown in Table 1,
complications and similar DCSI and CCI scores. Decrea- suggesting that the matching successfully created groups
sers had more PCP visits than did Non-decreasers, although that were similar with respect to observable characteristics.
both groups had a comparable number of endocrinologist
visits. Prior to matching, total healthcare costs during the 3.3 A1C Change
baseline period were also about $1000 higher, on average,
for patients who decreased A1C compared to patients who Prior to matching, the average first A1C was 10.8% among
did not decrease A1C (Table 1). Decreasers, 10.3% among Non-decreasers, and 10.1%
among patients whose A1C did not change. The average

Table 1 Baseline characteristics with and without propensity score matching

Variable Prior to matching* After propensity score matching*
Decreasers Non-decreasers p value Decreasers Non-decreasers p value
(N = 2273) (N = 924) (N = 912) (N = 912)

Female, N (%) 938 (41) 427 (46) 0.0104  407 (45) 421 (46) 0.5103
Age, mean, years (SD) 56.3 (12.3) 54.8 (11.8) 0.0014  55 (12) 55 (12) 0.8349
Long-term complications
None, N (%) 1247 (55) 505 (55) 0.9148 499 (55) 502 (55) 0.8877
Microvascular only (%) 383 (17) 173 (19) 0.2053 170 (19) 172 (19) 0.9045
Macrovascular only (%) 371 (16) 129 (14) 0.0957 135 (15) 129 (14) 0.6897
Microvascular and 272 (12) 117 (13) 0.5854 108 (12) 109 (12) 0.9423
macrovascular (%)
Region 0.5924
Northeast (%) 199 (9) 76 (8) 73 (8) 76 (8) 0.7976
West (%) 458 (20) 165 (18) 159 (17) 165 (18) 0.7132
Midwest (%) 194 (9) 80 (9) 76 (8) 80 (9) 0.7377
South (%) 1399 (62) 594 (64) 604 (66) 591 (65) 0.5219
Unknown (%) 23 (1) 9 (1)
CCI, mean (SD) 4.3 (2.3) 4.2 (2.2) 0.1737 4.2 (2.2) 0.4939
DCSI, mean (SD) 1.07 (1.5) 1.13 (1.5) 0.3637 1.0 (1.5) 1.1 (1.5) 0.3069
Calendar year 0.0053 
2009 615 (27) 308 (33) 327 (36) 306 (34) 0.3016
2010 616 (27) 242 (26) 231 (25) 237 (26) 0.7477
2011 398 (18) 152 (16) 136 (15) 148 (16) 0.4384
2012 366 (16) 119 (13) 109 (12) 118 (13) 0.5232
2013 278 (12) 103 (11) 109 (12) 103 (11) 0.6611
Number of endocrinologist visits, 0.40 (1.1) 0.40 (1.1) 0.7175 0.4 (1.1) 0.4 (1.1) 0.9493
mean (SD)
Number of PCP visits, mean (SD) 4.40 (3.3) 3.90 (3.3) 0.0003  3.9 (2.7) 3.9 (3.2) 0.9874
Total costs, mean (SD) $11,998 $10,691 ($14,760) 0.0693   $11,112 $10,642 ($14,733) 0.5330
($19,747) ($17,398)
First A1C, mean (SD) 10.8% (1.6) 10.3% (1.2) \0.0001  10.8% (1.6) 10.3% (1.2) \ 0.0001 
CCI Charlson Comorbidity Index, DCSI Diabetes Complication Serverity Index, PCP primary-care provider, A1C glycated hemoglobin
*
Decreaser and Non-decreaser cohorts compared using chi-square test for categorical variables and t tests for continuous variables
 
p \ 0.05
  
p \ 0.10
 M. Bansal et al.

change in A1C was -2.3 percentage points for Decreasers $14,500

$14,091
$14,223
and ?1.1 percentage points for Non-decreasers. Of the

Mean healthcare costs per year per

$14,000

Decreasers, 73% had an A1C that decreased by 1 per- $13,500

centage point or more, with an average change in A1C of - $13,000 $13,004

paent
2.9 percentage points (Table 2). $12,500

$12,000
$12,059
3.4 Total Healthcare Costs $11,500 $11,112
Decreasers (N=912)
$11,000
$10,641 Non-decreasers (N=912)
When matched patients were followed for 2 years, both $10,500

groups had increased total costs. However, the magnitude $10,000

Baseline 1st year post Index* 2nd year post Index
of the cost increase between the baseline and follow-up
periods was significantly lower among the Decreasers than Fig. 3 Mean total cost per patient per year. *Costs in first year of
among the Non-decreasers (Fig. 3). follow-up are statistically significant at p \ 0.05 compared to
baseline costs
As shown in Table 3, DD analysis indicated that, during
the first year of follow-up, Decreasers spent $2503 less on
a one percentage point decrease in A1C, compared to
total healthcare and $2123 less on medical costs, relative to
patients who had less than one percentage point decrease in
Non-decreasers (p \ 0.10). Similarly, during the second
A1C; however, this difference between the subgroups was
year of follow-up (Table 4), Decreasers experienced a
not statistically significant (Table 6). Furthermore, the
$1690 per patient per year reduction in total healthcare
individual subgroups did not show a difference in growth
costs (p \ 0.10) and a $1236 per patient per year reduction
in spending between Decreasers and Non-decreasers. The
in medical costs during the baseline period, relative to
regression coefficients for each subgroup along with the
Non-decreasers (p \ 0.10). In other words, patients whose
triple interaction and p-values can be found in Online
A1C decreased had average annual healthcare costs that
Supplementary Material, Appendix 3.
were 24% lower during the first year of follow-up and 17%
lower during the second year of follow-up, compared to
patients whose A1C increased/did not change during the
4 Discussion
same time periods. In both time periods, the outpatient
category was the largest contributor to cost savings ($1597
Diabetes is a progressive and chronic disease that is typi-
and $743 per patient per year, respectively), although this
cally associated with increased healthcare costs over time.
was not statistically significant (Tables 3, 4).
Uncontrolled diabetes is associated with even greater
Descriptive analysis of outpatient costs from the mat-
healthcare costs, because it leads to complications, such as
ched cohorts indicated that cost savings for Decreasers
coronary artery disease, stroke, nephropathy, neuropathy,
occurred primarily within two categories, diagnostics and
and retinopathy [2]. Our analysis of a nationally repre-
ancillary services related to DME (diabetes and non-dia-
sentative population with uncontrolled diabetes suggests
betes related). Decreasers realized additional, albeit smal-
that A1C reduction is associated with slower growth in
ler, cost savings in all other outpatient categories, except
total healthcare costs, an effect observed throughout a
the Other Services category (Table 5).
2-year follow-up period. More specifically, patients whose
When we repeated the propensity score matching for
A1C decreased had average annual healthcare costs that
two sub-groups of Decreasers compared to Non-decreasers
were 24% lower during the first year of follow-up and 17%
and conducted a test of the triple difference, we found
lower during the second year of follow-up, compared to
slower growth in costs of $670 for patients who had at least

Table 2 First A1C, index A1C, and change in A1C within the study groups
Patients with first A1C C 9% (N = 3197) First A1C (mean, SD) Index A1C (mean, SD) D A1C (mean)

Decreased A1C N = 2273 (71%) 10.8 (1.6) 8.5 (1.6) - 2.3

By C 1% N = 1655 (73%) 10.9 (1.7) 8.1 (1.5) - 2.9
By \1% N = 618 (27%) 10.3 (1.2) 9.8 (1.2) - 0.5
Increased A1C N = 861 (27%) 10.3 (1.2) 11.4 (1.6) 1.1
No change in A1C N = 63 (2%) 10.1 (1.26) 10.1 (1.26) 0
A1C glycated hemoglobin
Impact of Reducing Glycated Hemoglobin on Healthcare Costs

Table 3 Analysis of cost categories during the first year of follow-up

Cost Cost at baseline (mean) per Cost at 1st year post-index (mean) per Difference p value
category patient per year patient per year

Totala Decreasers $11,112 $12,059 $947 0.0397  

Non-decreasers $10,641 $14,091 $3450
1st year cost difference-in- - $2,503
difference
Medicalb Decreasers $7245 $7771 $525 0.0701 
Non-decreasers $6509 $9157 $2649
1st year cost difference-in- - $2123
difference
Inpatient Decreasers $1134 $1129 - $5 0.1541
Non-decreasers $923 $1428 $505
1st year cost difference-in- - $510
difference
Outpatient Decreasers $5983 $6509 $526 0.1353
Non-decreasers $5413 $7536 $2122
1st year cost difference-in- - $1597
difference
ER Decreasers $128 $133 $5 0.5687
Non-decreasers $173 $194 $21
1st year cost difference-in- - $17
difference
Pharmacyc Decreasers $3824 $4249 $425 0.0573 
Non-decreasers $4118 $4831 $713
1st year cost difference-in- - $288
difference
ER emergency room
 
Statistically significant at p \ 0.05
  
Statistically significant at p \ 0.10
a
Total Cost = Inpatient ? Outpatient ? Emergency Room (ER) ? Pharmacy ? Insulin Pump and Blood Glucose Meter Costs
b
Medical Costs = Inpatient ? Outpatient ? ER Costs
c
Pharmacy = All Drugs ? Supplies Cost

patients whose A1C increased/did not change, during the Hawaiian health insurance plan who had at least a 2-per-
same time periods. Other studies have observed similar centage point decrease in A1C (i.e., from [9% to \7%)
trends in total healthcare cost savings; however, the mag- and sustained control over a 3-year follow-up period. In
nitude of the savings was relatively low and differences in contrast to results in the current study, the investigators
methodology make direct comparisons difficult [14, 18]. found that reducing A1C to \7% was not associated with
Research by Wagner et al. and Juarez et al. was similar cost reductions in any given year of the study. However,
to that of the present study, in that both focused on the sustained A1C control was associated with a savings of
relationship between changes in patients’ A1C levels and more than $5000 over the course of the 3-year follow-up
changes in healthcare costs [12, 16]. In the former study, period [16].
the sample was selected from a staff-model HMO in Several other retrospective analyses incorporated a
Western Washington in the mid-1990s. The investigators cross-sectional design to demonstrate healthcare cost sav-
found that a sustained decrease in A1C of at least one ings for patients with good versus poor A1C control over
percentage point was associated with healthcare cost sav- the short term. However, unlike our protocol, ‘‘short-term’’
ings within 1–2 years of improvement; however, the dif- was defined as a 3- to 5-year follow-up period
ferences were statistically significant only for those whose [13, 15, 17, 18]. An exception is a study by Shetty et al.,
baseline A1C was C 10% [12]. The latter study had a which included a 1-year follow-up. Their findings sug-
similar design, but focused on patients enrolled in an gested that total diabetes-related costs were significantly
 M. Bansal et al.

Table 4 Analysis of cost categories during second year of follow-up

Cost Cost at baseline (mean) per Cost at 2nd year post-index (mean) per Difference p value
category patient per year patient per year

Totala Decreasers $11,112 $13,004 $1892 0.1674

Non-decreasers $10,641 $14,223 $3581
2nd year cost difference-in- - $1690
difference
Medicalb Decreasers $7245 $8388 $1142 0.2552
Non-decreasers $6509 $8887 $2378
2nd year cost difference-in- - $1236
difference
Inpatient Decreasers $1134 $1474 $339 0.2393
Non-decreasers $923 $1739 $816
2nd year cost difference-in- - $476
difference
Outpatient Decreasers $5983 $6741 $758 0.4298
Non-decreasers $5413 $6914 $1501
2nd year cost difference-in- - $743
difference
ER Decreasers $128 $173 $45 0.9937
Non-decreasers $173 $234 $61
2nd year cost difference-in- - $16
difference
Pharmacyc Decreasers $3824 $4548 $724 0.3282
Non-decreasers $4118 $5236 $1118
2nd year cost difference-in- - $394
difference
ER emergency room
 
Statistically significant at p \ 0.05
  
Statistically significant at p \ 0.10
a
Total cost = Inpatient ? Outpatient ? Emergency Room (ER) ? Pharmacy ? Insulin Pump and Blood Glucose Meter Costs
b
Medical costs = Inpatient ? Outpatient ? ER Costs
c
Pharmacy = All drugs ? Supplies Costs

higher for those with poor A1C control (A1C [7%) comfort between the primary-care physician and the
compared to those with A1C B 7% [14]. However, the patient, resulting in less physician visits [12].
study included only patients with type 2 diabetes and did Improvements in glycemic control may also provide
not examine the effect of reducing A1C over time. positive reinforcement for a patient’s efforts in managing
When the distribution of cost savings across healthcare his or her illness, which may subsequently increase self-
cost categories was analyzed, the greatest savings occurred efficacy and reduce dependency on medical care for dia-
in the outpatient category at each of the follow-up time betes management [12]. A retrospective, cross-sectional
points, with nearly 40% of the savings observed within the claims analysis by Degli Esposti et al. also found a positive
diagnostic testing subcategory. This suggests that Non- association between A1C levels and both total and outpa-
decreasers may have used more diagnostic services during tient healthcare costs over a 2-year period; however, the
the follow-up period. Because the cost savings observed in study focused only on Italian patients with type 2 diabetes
the current study occurred within a short timeframe, it was [17].
unlikely to be related to complication prevention [12]. In addition to studying total healthcare costs, a number
Instead, reduced outpatient healthcare costs among patients of healthcare cost categories (e.g., inpatient, outpatient,
with better glycemic control may be related to symptomatic etc.), as well the subgroup of patients whose A1C
relief and improvements in quality of life [23]. Wagner decreased C 1 percentage point compared to patients who
et al. suggest that better glycemic control may increase the decreased \ 1 percentage point were examined. Regarding
Impact of Reducing Glycated Hemoglobin on Healthcare Costs

Table 5 Analysis of outpatient costs

Outpatient categories Cost difference-in-difference Proportion of outpatient cost saving
(%)

Diagnostics (Laboratory, X-rays, pathology, allergy tests etc.) - $632 (lower for 39.8
Decreasers)
Ancillary services related to durable medical equipment - $396 (lower for 24.9
Decreasers)
Outpatient, Other - $67 (lower for 4.2
Decreasers)
Office Visits - $268 (lower for 16.9
Decreasers)
Outpatient ER - $124 (lower for 7.8
Decreasers)
Outpatient Surgery - $79 (lower for 5.0
Decreasers)
Other Services (Physical therapy, home health, vision, mental health - $21 (lower for 1.3
therapy etc.) Decreasers)
Total - $1587 (lower for
Decreasers)
ER emergency room

Table 6 First-year difference-in-difference (DD) by subgroup of decreasers and triple difference

Subgroup of decreasers Individuals in each group Total costs Medical costs Pharmacy costs
DD, 1st Year p value DD, 1st Year p value DD, 1st Year p value

By magnitude of decrease
A1C decreased by C 1% 909 - $992 0.4184 - $648 0.5693 - $223 0.1156
A1C decreased by \1% 584 - $323 0.7852 - $86 0.963 - $184 0.2536
Triple difference - $670 0.9327 - $563 0.8932 - $39 0.7059

the latter group, patients whose A1C decreased by 1 per- stable, or increase during the follow-up period was not
centage point or more experienced more cost savings than evaluated, because sample sizes were too small.
did patients who decreased their A1C by less than 1 per- Despite these limitations, results suggest that patients
centage point, in line with expectations. However, this with uncontrolled diabetes whose A1C decrease experience
difference was not statistically significant. relatively slower growth in total healthcare costs than do
This study had several limitations that must be taken patients with uncontrolled diabetes whose A1C does not
into account when interpreting the findings. The first lim- decrease. However, unlike the study by Gilmer et al.,
itation is that the Optum Clinformatics dataset does not which identified significant healthcare cost savings 5–10
include lab data for all patients, which could cause selec- years following A1C reduction, and attributed those sav-
tion bias. In fact, a single A1C value was available for only ings to the positive effect on complications [13], this study
30% of the population sample, and two A1C values were identified cost savings within 1–2 years of decreased A1C,
available for less than 30%. The second is that the costs in which supports findings by Shetty et al. [14].
the Optum Clinformatics data set are standardized and do Given our results, it seems important to focus on helping
not reflect the allowed amount an insurance plan will pay patients with diabetes achieve A1C control. Previous
for a covered healthcare service. Third, the propensity research has shown that intervention programs that facili-
score method did not include matching on type of diabetes, tate early diagnosis of diabetes, regular monitoring of each
type of diabetes therapy, number of different medications patient’s progress, patient education, and systematic fol-
prescribed, duration of diabetes, or race/ethnicity, income, low-up will help individuals attain glycemic goals and will
or education, all of which may have impacted results. ultimately result in healthcare costs savings [4, 24].
Finally, whether A1Cs continued to decrease, remain
 M. Bansal et al.

Our findings are useful for policymakers, payers, and 7. American Diabetes Association. 6. Glycemic Targets. Diabetes
others interested in moving from fee-for-service to value- Care. 2017;40(Suppl 1):S48–56.
8. Disease Management Performance Measures. 2017. http://www.
based reimbursement for diabetes-related healthcare. In ncqa.org/programs/accreditation/disease-management-dm/dm-
addition, initiatives that target uncontrolled diabetes by performance-measures. Accessed 1 Nov 2017.
reducing A1C levels may lead to cost savings within a rel- 9. Comprehensive Diabetes Care. 2017. http://www.ncqa.org/
atively short time period, providing an incentive for stake- report-cards/health-plans/state-of-health-care-quality/2016-table-
of-contents/diabetes-care. Accessed 17 Nov 2017.
holders to engage in value-based reimbursement, which will 10. Eastman RC, Javitt JC, Herman WH, Dasbach EJ, Zbrozek AS,
reduce the burden on the overall healthcare system over time. Dong F, et al. Model of complications of NIDDM. I. Model
construction and assumptions. Diabetes Care.
Acknowledgements The authors thank Keren Price, MS, RD, for her 1997;20(5):725–34.
assistance with writing this manuscript, and Toni Cordero, PhD, for 11. Eastman RC, Javitt JC, Herman WH, Dasbach EJ, Copley-Mer-
her assistance with manuscript review. riman C, Maier W, et al. Model of complications of NIDDM. II.
Analysis of the health benefits and cost-effectiveness of treating
Authors’ contributions Megha Bansal—developed concept, con- NIDDM with the goal of normoglycemia. Diabetes Care. 1997
ducted statistical analysis, developed manuscript. Mona Shah—de- May;20(5):735–44.
veloped concept, reviewed statistical analysis and manuscript. Brian 12. Wagner EH, Sandhu N, Newton KM, McCulloch DK, Ramsey
Reilly—provided data by extracting it from larger database and SD, Grothaus LC. Effect of improved glycemic control on health
reviewed manuscript. Susan Willman—reviewed concept, results, and care costs and utilization. JAMA. 2001;285(2):182–9.
manuscript. Max Gill—developed the concept; reviewed statistical 13. Gilmer TP, O’Connor PJ, Rush WA, Crain AL, Whitebird RR,
analysis and manuscript. Francine R. Kaufman—provided a clinical Hanson AM, et al. Predictors of health care costs in adults with
perspective and reviewed manuscript. diabetes. Diabetes Care. 2005;28(1):59–64.
14. Shetty S, Secnik K, Oglesby AK. Relationship of glycemic
Compliance with Ethical Standards control to total diabetes-related costs for managed care health
plan members with type 2 diabetes. J Manag Care Pharm.
Data availability statement According to our contract with 2005;11(7):559–64.
OptumInsight, we are not permitted to share data from the Optum 15. Menzin J, Korn JR, Cohen J, Lobo F, Zhang B, Friedman M,
ClinformaticsÒ Data Mart database outside of our organization. We et al. Relationship between glycemic control and diabetes-related
have provided details of the model in an Appendix (see Online hospital costs in patients with type 1 or type 2 diabetes mellitus.
Supplementary Material). J Manag Care Pharm. 2010;16(4):264–75.
16. Juarez D, Goo R, Tokumaru S, Sentell T, Davis J, Mau M.
Ethical standards All authors (Megha Bansal, Mona Shah, Brian Association between sustained glycated hemoglobin control and
Reilly, Susan Willman, Max Gill, and Francine R. Kaufman) are healthcare costs. Am J Pharm Benefits. 2013;5(2):59–64.
employees of Medtronic. Medtronic has an interest in this paper 17. Degli Esposti L, Saragoni S, Buda S, Sturani A, Degli Esposti E.
because it illustrates the impact of decreasing versus increasing A1C Glycemic control and diabetes-related health care costs in type 2
on short-term costs in patients with uncontrolled diabetes. diabetes; retrospective analysis based on clinical and adminis-
trative databases. Clinicoecon Outcomes Res. 2013
May;14(5):193–201.
Funding No financial assistance was used to conduct this study or
18. Oglesby AK, Secnik K, Barron J, Al-Zakwani I, Lage MJ. The
prepare this manuscript.
association between diabetes related medical costs and glycemic
control: a retrospective analysis. Cost Eff Resour Alloc.
2006;4:1-7547-4-1.
19. Menzin J, Langley-Hawthorne C, Friedman M, Boulanger L,
References Cavanaugh R. Potential short-term economic benefits of
improved glycemic control: a managed care perspective. Diabetes
Care. 2001;24(1):51–5.
1. Centers for Disease Control and Prevention. National Diabetes
Statistics Report. 2017. Atlanta, GA: Centers for Disease Control 20. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of
and Prevention, US Dept of Health and Human Services, 2017. classifying prognostic comorbidity in longitudinal studies: devel-
opment and validation. J Chronic Dis. 1987;40(5):373–83.
2. American Diabetes Association. Economic costs of diabetes in
21. Clinformatics Data Mart. 2018. https://www.optum.com/content/
the US in 2012. Diabetes Care. 2013;36(4):1033–46.
3. Boyle JP, Thompson TJ, Gregg EW, Barker LE, Williamson DF. dam/optum/resources/productSheets/Clinformatics_for_Data_
Projection of the year 2050 burden of diabetes in the US adult Mart.pdf. Accessed 30 Apr 2018.
22. Young BA, Lin E, Von Korff M, Simon G, Ciechanowski P,
population: dynamic modeling of incidence, mortality, and pre-
Ludman EJ, et al. Diabetes complications severity index and risk
diabetes prevalence. Popul Health Metr. 2010;8:29-7954-8-29.
4. American Diabetes Association. Standards of medical care in of mortality, hospitalization, and healthcare utilization. Am J
Manag Care. 2008;14(1):15–23.
diabetes-2017 abridged for primary care providers. Clin Diabetes.
23. Testa MA, Simonson DC. Health economic benefits and quality
2017;35(1):5–26.
5. Diabetes Control and Complications Trial Research Group, of life during improved glycemic control in patients with type 2
Nathan DM, Genuth S, Lachin J, Cleary P, Crofford O, et al. The diabetes mellitus: a randomized, controlled, double-blind trial.
JAMA. 1998;280(17):1490–6.
effect of intensive treatment of diabetes on the development and
24. Schwartz DS, Dick DE, McMahon C, Cohen DO, Knoulton J,
progression of long-term complications in insulin-dependent
diabetes mellitus. N Engl J Med. 1993 Sep 30;329(14):977–86. Arunachalam S. Novel medtronic turning point program
6. Herman WH, Eastman RC. The effects of treatment on the direct improves compliance and HBA1C in at-risk patients with type 1
or type 2 diabetes. Diabetes Technol Ther. 2017;19(S1):A59.
costs of diabetes. Diabetes Care. 1998 Dec;21(Suppl 3):C19–24.