Costs and Outcomes Comparison of Diabetes Technology Usage Among People With Type 1 or 2 Diabetes Using Rapid-Acting Insulin

Abstract

Background:

Does initiation of a continuous glucose monitor (CGM) or insulin pump lower health care utilization and/or costs?

Methods:

Distinct cohorts of people with type 1 diabetes (T1D) or type 2 diabetes (T2D) using a blood glucose monitor (BGM), CGM, pump, or CGM with pump were identified from a large claims database. Patients ≥40 years old with 12 months of continuous enrollment before and after the device start date qualified for the study. Outcomes included one-year medical utilization and costs (minus device) for events such as hospitalizations and office visits. Generalized linear models were fitted, controlling for numerous baseline covariates. The Holm method corrected for the multiplicity of hypotheses tested.

Results:

Of the 8235 total patients, the BGM control group was the largest, had the lowest percentage of patients with T1D, and was significantly different from the device groups in most baseline categories. Formally, only two comparisons were statistically significant: Compared with BGM, the pump cohort had greater adjusted first-year total medical and office visit costs. Other secondary outcomes such as days hospitalized, emergency department visits and labs, favored pump. Most endpoints were favorable for CGM. Results for CGM with pump generally were intermediate between CGM and pump alone.

Conclusions:

During a one-year follow-up, unadjusted medical costs of both CGM and pump appear lower than BGM, but multivariable modeling yielded adjusted savings only for CGM use. Economic benefits might be observable sooner for CGMs than for pumps. Generalized linear models fitted to health care utilization event rates produced favorable results for both CGM and pump.

Keywords

continuous glucose monitor generalized linear model health economics insulin pump medical costs real-world study

Introduction

Diabetes is one of the most prevalent chronic conditions in the United States, with 10.5% of the population, or 34.2 million people, affected in 2018.¹ The economic cost of diabetes in the United States was estimated to be $327 billion in 2017, including $237 billion in direct medical costs.² These costs are increasing every year due to the growing prevalence of type 2 diabetes (T2D) and the rising costs per person with diabetes.

Diabetes technology, such as continuous subcutaneous insulin infusion (CSII, or insulin pump), continuous glucose monitoring (CGM) devices, and hybrid closed-loop systems, may improve diabetes outcomes but at an increased cost. Studies estimating the incremental costs have yielded intriguing but inconclusive results.^3-6 For example, the populations have included only type 1 diabetes⁵ (T1D) or T2D,^3,4,6 only one type of device,^4,5 excluded a control group,³ or derived the results from a model.⁶ We sought to overcome these limitations by including a more comprehensive population (ie, CGMs and pumps, both T1D and T2D) and a rigorous methodology (multivariable, longitudinal analysis with a control group).

In this study, we used real-world evidence from a large claims database to identify distinct cohorts of people with T1D or T2D who used a blood glucose monitor (BGM), CGM, pump, or CGM with pump. We compared these groups for all-cause and diabetes-related medical costs and outcomes such as hospitalizations, labs, and office visits by fitting generalized linear models (GLMs) that controlled for an abundance of baseline covariates.

Methods

Patients and Study Design

This retrospective observational study utilized insurance claims from the IBM MarketScan Commercial and Medicare Supplemental Databases, which provides detailed cost, use, and outcomes data for health care services performed in both inpatient and outpatient settings. The medical claims are linked to outpatient prescription drug claims and person-level enrollment data using unique enrollee identifiers. All database records are de-identified and compliant with US patient confidentiality requirements, including the Health Insurance Portability and Accountability Act of 1996, and thus institutional review board approval was not necessary for this study.

Patients were included who used CSII (pump), CGM, or BGM during 2015-2017 (n = 266 699); had at least two diagnoses or claims indicating T1D or T2D during the 12-month baseline (n = 209 438); had at least 12 months of continuous enrollment before and after the index date (defined below, n = 89 978); were at least 40 years old on the index date (n = 76 790); remained on the pump, CGM or BGM for at least one year (n = 22 288); and received two or more prescriptions for rapid-acting insulin during baseline (n = 8235).

The index date was defined as the first claim for CGM or pump between January 1, 2015, and December 31, 2017, with no prior claim for CGM or pump during the previous 6 months. If a patient had start dates for a CGM and a pump within 30 days of each other, the patient qualified for the “CGM & pump” cohort; all others with both pump and CGM use were excluded.

Patients with a claim for BGM between July 1, 2014, and December 31, 2017, who were not identified above as potential CGM or pump patients, formed the BGM cohort. To assign an index date to BGM patients, a simple randomization procedure was used to ensure that the distribution of time between the first BGM claim date and the index date was roughly similar in the BGM and CGM groups.

This scheme produced four distinct and mutually exclusive cohorts of patients: BGM (no pump or CGM), CGM (no pump), pump (no CGM), and CGM & pump (Figure 1). We compared these groups with respect to baseline characteristics, diabetes outcomes, and all-cause and diabetes-related costs in the 12-month follow-up period.

Figure 1.

Study design with mutually exclusive cohorts.

Statistical Analysis

About 30 relevant baseline covariates were measured for each patient: demographics (eg, age, sex, region, and type of health insurance), overall health scores (eg, Charlson Comorbidity score^7,8 and polypharmacy, the count of distinct medication classes), and others. The following all-cause and diabetes-related medical costs during the follow-up were separately modeled as the response:

Hospitalizations,

Emergency department (ED) visits,

Office visits,

Labs, and

Total medical (ie, nonpharmacy, except device/supply costs).

Device costs, including supplies, were calculated and reported separately. Although pharmacy costs were not included in the analysis, National Drug Codes were part of the code list used to identify both pumps and CGMs (see Supplemental Material). The gross payment to the provider after applying discounts—but before applying deductibles, copayments, and so on—was used to determine costs. All costs were adjusted to 2018 costs using health care inflation rates.⁹

A GLM with a gamma distribution and log link was fitted to the cost endpoints, adjusting for all covariates.¹⁰ Because the gamma distribution only admits nonnegative values, the formal dependent variable was the one-year cost during follow-up. By always including the baseline cost as a covariate, the model effectively considered the change in medical costs (excluding device costs) from the year prior to the year after the index date. Model results compare the difference, or more accurately, the ratio of follow-up costs between the BGM control group and each of the device groups, informed by the baseline cost and all other covariates. Backward variable reduction was applied to retain in the model only significant predictors (P < .05).

To determine the significance of the multiple primary comparisons while safeguarding against multiplicity, we applied the method outlined by Holm,¹¹ which assigns α/k significance level to compare with the smallest of k P-values, α / (k – 1) to the second smallest, and so on, until the first nonrejection, when the procedure stops. For the purposes of multiple testing, all-cause comparisons were considered the primary comparisons and diabetes-related secondary. Also, pump comparisons were considered primary while CGM was secondary. As a sensitivity analysis for the primary endpoint, costs during the 12-month follow-up period were analyzed using a GLM with inverse probability of treatment (IPT) weights.¹²

Health care utilization event counts were modeled by fitting a GLM with a Poisson distribution and log link.¹³ In addition to the outcomes listed above, total days of hospitalization and endocrinologist office visits were also modeled as event outcomes, whereas no equivalent to total medical was fitted, for a total of six endpoints.

Results

Study Population

A total of 8235 patients fit the inclusion criteria over the three index years of 2015-2017. The demographic and baseline clinical characteristics of the study population, by cohort, appear in Table 1. The BGM control group was the largest and oldest cohort, had the lowest percentage of patients with T1D, and generally had the highest frequency of complications and medication use. The CGM wearers included the largest percentage of men and the most residents of the Northeast and West regions. Pump users featured the most Southern and the least Western residents, with the most retinopathy and neuropathy among the four cohorts, but the least hyperglycemia. The smallest group was the dual wearers of both a CGM and a pump, by and large the youngest and healthiest of the four cohorts and with the highest percentage of T1D patients.

Table 1.

Demographics and Baseline Characteristics.

	BGM(N = 5155)		CGM(N = 966)		P value	Pump(N = 1579)		P value	CGM & pump(N = 535)		P value
	n	% or mean ± SD	n	% or mean ± SD	[1]	n	% or mean ± SD	[1]	n	% or mean ± SD	[1]
Index year
2015	2013	39.0	273	28.3	<.0001	950	60.2	<.0001	261	48.8	<.0001
2016	1663	32.3	272	28.2		353	22.4		157	29.3
2017	1479	28.7	421	43.6		276	17.5		117	21.9
Female	2283	44.3	407	42.1	.22	819	51.9	<.0001	252	47.1	.21
Age (years)	5155	59.5 ± 10.2	966	54.6 ± 8.7	<.0001	1579	57.3 ± 9.9	<.0001	535	52.1 ± 7.5	<.0001
40-49	856	16.6	303	31.4	<.0001	391	24.8	<.0001	208	38.9	<.0001
50-59	1861	36.1	391	40.5		569	36.0		230	43.0
60+	2438	47.3	272	28.2		619	39.2		97	18.1
Type 1 DM (vs T2)	943	18.3	599	62.0	<.0001	909	57.6	<.0001	369	69.0	<.0001
Health care payer type
Commercial (vs Medicare)	3856	74.8	853	88.3	<.0001	1206	76.4	.20	516	96.4	<.0001
Geographic region
Northeast	859	16.7	213	22.0	<.0001	323	20.5	<.0001	92	17.2	.0003
Midwest	1394	27.0	174	18.0		345	21.8		127	23.7
South	2298	44.6	364	37.7		756	47.9		220	41.1
West	604	11.7	215	22.3		155	9.8		96	17.9
Charlson Comorbidity Score	5155	3.5 ± 2.4	966	3.0 ± 2.2	<.0001	1579	3.1 ± 2.1	<.0001	535	2.9 ± 1.9	<.0001
Depression	741	14.4	116	12.0	.052	218	13.8	.57	70	13.1	.42
Anxiety	508	9.9	81	8.4	.16	139	8.8	.21	53	9.9	.97
Potential diabetes-related complications
Cardiovascular events	1598	31.0	188	19.5	<.0001	400	25.3	<.0001	72	13.5	<.0001
Cerebrovascular events	360	7.0	37	3.8	.0003	102	6.5	.47	21	3.9	.0071
Peripheral vascular events	824	16.0	96	9.9	<.0001	195	12.3	.0004	48	9.0	<.0001
Retinopathy	1265	24.5	317	32.8	<.0001	565	35.8	<.0001	168	31.4	.0005
Nephropathy	1417	27.5	179	18.5	<.0001	348	22.0	<.0001	100	18.7	<.0001
Neuropathy	1686	32.7	256	26.5	.0001	547	34.6	.15	149	27.9	.022
Hypoglycemic events	591	11.5	214	22.2	<.0001	280	17.7	<.0001	118	22.1	<.0001
Hyperglycemic events	253	4.9	53	5.5	.45	47	3.0	.0011	37	6.9	.044
Adapted Diabetes Complications Severity Index	5155	2.0 ± 2.1	966	1.5 ± 1.9	<.0001	1579	1.9 ± 2.0	.0163	535	1.5 ± 1.8	<.0001
Polypharmacy Index	5155	6.8 ± 2.7	966	5.9 ± 2.7	<.0001	1579	6.0 ± 2.7	<.0001	535	6.0 ± 2.8	<.0001
5+ meds	4045	78.5	635	65.7		1090	69.0		364	68.0
Diabetes-related treatment
Insulin: Basal	4773	92.6	922	95.4	.0014	557	35.3	<.0001	398	74.4	<.0001
Insulin: Bolus	5155	100	966	100	1	1579	100	1	535	100	1
Glucagon	178	3.5	159	16.5	<.0001	124	7.9	<.0001	96	17.9	<.0001
Other [2]	4482	86.9	529	54.8		709	44.9		277	51.8
HbA1c [3]	371	8.6 ± 2.0	90	8.4 ± 1.7	.24	81	8.5 ± 1.7	.44	42	8.7 ± 1.5	.96

[1] Unadjusted P values for device groups compared with control (BGM) group are from two-tailed Student t test for continuous measures and chi-square test or Fisher exact test if expected cell <5 for categorical measures. [2] Other included GLP-1, DPP-4, and other oral antidiabetic medications. [3] Reported HbA1c value for a limited subgroup: the value closest to or on index date. Baseline period: 12 months prior to index date; Index Period: January 1, 2015-December 31, 2017.

Abbreviations: BGM, blood glucose monitor; N, total cohort size; n, number of patients; SD, standard deviation; CGM, continuous glucose monitor; DM, diabetes mellitus; T2, type 2 diabetes; GLP-1, glucagon-like peptide 1; DPP-4, dipeptidyl peptidase-4; HbA1c, hemoglobin A1C.

The raw average annual health care costs for the four groups during the baseline and follow-up periods appear in Table 2. For BGM, total medical costs decreased slightly from baseline to follow-up, while they increased for CGM and pump. The unadjusted mean differences and the model-adjusted differences (least squares means) are shown in Figure 2.

Table 2.

Average Annual Health Care Costs (Raw), in 2018 US Dollars.

	BGM	CGM	Pump	Pump & CGM
N	5155	966	1579	535
Total medical
Baseline	$25 410	$17 461	$17 585	$17 084
Follow-up	$24 633	$18 424	$20 705	$15 485
Hospitalizations
Baseline	$10 477	$5685	$5181	$7565
Follow-up	$9272	$5233	$6799	$4337
Emergency department
Baseline	$766	$699	$667	$681
Follow-up	$795	$706	$734	$815
Physician office
Baseline	$3336	$3979	$3235	$2804
Follow-up	$3508	$4193	$3637	$3227
Labs
Baseline	$1153	$704	$595	$886
Follow-up	$1001	$744	$818	$728

Abbreviations: BGM, blood glucose monitor; CGM, continuous glucose monitor; N, total cohort size.

Figure 2.

Raw and model-adjusted follow-up cost differences, device ‒ BGM.

Health care utilization for the four cohorts during the baseline and follow-up periods is shown in Figure 3. The CGM group had the lowest percentage of patients with at least one hospitalization, ED visit, or laboratory analysis during follow-up, whereas BGM had the highest frequency in all three outcomes. The situation was reversed for endocrinologist and primary care physician visits (not shown), with BGM having the lowest frequencies at follow-up compared with the devices.

Figure 3.

Health care resource utilization (% of patients with ≥1 event).

Pump Versus BGM

The pump versus BGM all-cause cost comparisons were preselected as primary, so the first task of the Holm method was to order the P values from the five hypothesis tests. The smallest P value, P < .0001 for physician office visits, met the threshold of .01 set by Holm.¹¹ This result favored BGM and yielded an adjusted mean cost difference of $386, larger than the $128 observed difference, also in BGM’s favor. The next smallest P value, P = .0039 for total medical, also led to a rejection of the hypothesis of equal costs. Here, apparent savings in follow-up costs from the pump of $3927 compared with BGM were reversed by the fully adjusted model to yield a $1786 difference in favor of BGM. The third smallest P value, P = .041 for labs, was not formally significant, as it was larger than the Holm threshold of .0167. Formal hypothesis testing hence ended with only the two significant results, both favoring BGM. As can be appreciated from Figure 2, lab costs also favored pump prior to model adjustment, as did the cost of hospitalizations (P = .43). The only exception, not shown in the figure, was ED visit costs, in which both the raw and adjusted costs favored the pump (P = .083).

In general, the pump group had apparent savings in raw costs that were overturned by the models, likely driven by the BGM cohort’s overall older age and poorer health. Key covariates in the fitted GLMs were baseline costs, polypharmacy, Charlson Comorbidity Index, diabetes complications, diabetes type, and age. We also found an interaction of age by diabetes type, which had an apparent cutpoint at age 60 and whose significance was consistently driven by the T1D under 60 group incurring substantially less costs. Region also played a prominent role, with the Midwest and South usually featuring lower costs than the West and Northeast. Sex was significant in most models, with males usually incurring higher costs.

The correlations among the parameter estimates were reasonable, with absolute values seldom above .3, whereas the goodness-of-fit statistics yielded P values that were seldom below .2, indicating that the models fit the data well. Diabetes-related outcomes yielded similar results. When device costs were considered, BGM’s advantage was further extended, as the pump cohort incurred an average of $4512 in follow-up device costs compared with $319 for BGM.

The sensitivity analysis, involving IPT weights, was performed on physician office visits, total medical—the two significant endpoints—and labs. After estimating propensity scores by fitting a logistic regression model, the IPT weights were derived and plotted, and standardized differences were checked for several covariates to verify that the cohorts were balanced. Finally, a weighted GLM was fitted with only baseline cost and the device indicator as covariates. All three models yielded parameter estimates that agreed in size and directionality with the original parameter estimates, confirming the original results.

The pump fared much better in the event rate comparisons (Figure 4), although the test results cannot formally be called significant because the alpha had been spent. The models verified the pump group’s advantage in raw event rate with respect to days of hospitalization (P < .0001), ED visits (P = .0015), and labs (P = .0017), whereas BGM confirmed its unadjusted rate advantage in physician (P < .0001) and endocrinologist office visits (P < .0001). The one exception was hospitalizations, in which pump’s raw rate advantage (0.8 ratio with BGM) was erased by the model, yielding an adjusted rate ratio of 1.03 (P = .60), likely due to BGM’s hospitalization rate decreasing from baseline while pump's hospitalization rate increased. The relevant predictors were essentially the same as in the cost models, although age and gender dropped out of some models.

Figure 4.

Adjusted and unadjusted ratios of follow-up event rates, device ÷ BGM.

CGM Versus BGM

For this set of comparisons, the models typically reinforced the differences observed in unadjusted data. Had CGM versus BGM cost differences been the primary hypothesis tests, four P values would have been declared significant, all but one in favor of CGM; the exception was physician office visits (P = .017). For hospitalizations (P < .001), labs (P < .001), and, most importantly, total medical costs (P = .0076), the CGM group’s adjusted costs were lower. Prominent covariates in the GLMs were similar to the pump models. The lower costs with CGM are substantially offset if device costs are added, with an average of $4431 during the 12-month follow-up versus $319 for BGM.

Event rate comparisons mirrored the cost results. For CGM, lower rates were observed for hospitalizations (P < .0001), days of hospitalization (P < .0001), ED visits (P = .0010), and labs (P < .0001), whereas physician (P < .0001) and endocrinologist office visits (P < .0001) were lower for BGM. The dominant predictors followed the same pattern as in the cost models, with baseline events playing a preponderant role.

For both cost and event models, diabetes-related outcomes yielded similar results.

CGM & Pump Versus BGM

The cost comparisons for CGM & pump resembled the pump cohort’s, with the model reversing CGM & pump’s actual cost advantage in all but one outcome, so that the adjusted costs favored BGM. The P values, however, were orders of magnitude higher, giving an overall impression of cost equivalence, due to the CGM & pump group’s sample size being about a third of the pump’s sample size. Like CGM or pump alone, physician office visits (P = .0005) were higher compared with BGM. Age dropped out of every CGM & pump cost model, suggesting that cost is independent of age, at least for patients above 40 years. Device costs for CGM & pump were $9914.

The event rate models also mirrored the pump’s results, but with slightly larger P values. The CGM & pump group only achieved a notable effect with respect to days of hospitalization (P < .0001), whereas BGM confirmed its dominance over the devices in physician (P < .0001) and endocrinologist office visits (P < .0001).

Discussion

Interpretation of Results

With CGMs, economic benefits were apparent in the first year of use. Importantly, both hospitalizations and ED utilization were lower. Office visits were the only category of health care services that were higher after initiating CGM. Increased provider/person with diabetes (PwD) interactions are expected to accompany the need to interpret the extensive glucose data and modify management plans accordingly. If CGM and routine office visits can indeed reduce some avoidable acute care, a corresponding reduction in the growth of diabetes health care expenditures would be expected.

For the pump cohort, we observed excess adjusted costs for several endpoints, including the prespecified primary analyses. If our incremental cost estimates reasonably reflect the effect of a pump initiation, the value will depend upon improved clinical outcomes (eg, improved glycemic control leading to fewer complications) assessed over longer follow-up times. However, we also found that the rates of ED visits, hospital length of stay, and laboratory procedures were lower with pump use, although pump featured lower raw costs that the models could not confirm as significant. This apparent contradiction could reflect the complexity of the clinical situation and/or the analytical challenges with a real-world outcomes study, such as a nonrandomized control group and data limitations. Even with these potential issues, our results for both pump and CGM are consistent with previous randomized and real-world studies, as discussed below.

The cohort using both pump and CGM yielded results generally intermediate between either device alone, which suggested that the effects of the two devices were additive. The older generation of devices available in the current study period have been updated with more advanced versions of pump/CGM combinations (eg, hybrid closed loops). Whether these newer devices have synergistic effects could not be tested in this study.

By design, our study provided estimates of one-year economic outcomes associated with CGM and/or pump. Therefore, our results are not comprehensive assessments of potential cost savings or cost-effectiveness. The cost of the devices and related supplies would need to be considered as part of the overall budget impact; for example, CGM costs exceeded the savings derived from our models. In addition, although we did include some health care resource utilization outcomes, measuring the overall effectiveness was not feasible in this retrospective study. Some of the possible benefits, such as reductions in nonsevere hypoglycemia and improvements in quality of life (QoL), were not accessible. Finally, more than one year is required to measure long-term outcomes, such as those associated with the complications of diabetes.

Limitations, such as short follow-up time,⁴ data from more than a decade ago,³ and small sample sizes¹⁴ complicate the interpretation of prior literature. With those caveats in mind, comparisons of our results to previous real-world or pragmatic randomized studies are informative. For CGMs, evidence has mostly been favorable, such as fewer all-cause inpatient admissions and ED visits for diabetic ketoacidosis, compared with BGM.¹⁴ Another study found lower health care costs and fewer hospital admissions for CGM users.⁵ Similarly, flash glucose monitoring was associated with reductions in acute diabetes-related events and all-cause inpatient hospitalizations.¹⁵ Professional CGM, typically used for two weeks or less, was associated with $3776 lower annual costs in people with T2D.¹⁶ In a cost-effectiveness analysis as part of a randomized study, CGM was cost-effective in adults with T1D and suboptimal glycemic control, at the willingness-to-pay threshold of $100 000 per quality-adjusted life year.¹⁷ Outside the United States, there has been additional evidence generated, such as in Belgium, where diabetes-related hospitalizations decreased in the year after reimbursement was provided for CGM.¹⁸

Published results for insulin pumps have yielded more complicated outcomes, consistent with the mixed findings from our study. For example, analysis of data from a large US payer demonstrated substantially higher costs with pumps compared with injections.¹⁹ An extensive, long-term (five-year average follow-up) real-world study, from the Swedish National Diabetes Register, calculated annual mean excess costs for pump therapy equivalent to $3900 US dollars.²⁰ A cluster-randomized study in the United Kingdom concluded that pumps were not cost-effective but led to greater improvement in diabetes-specific QoL and treatment satisfaction.²¹ One of the least favorable results for insulin pumps came from adults with T1D who were already using CGM, which was associated with higher costs and reduced QoL.²² In contrast, long-term analyses from health economic models have typically shown pumps to be cost-effective.²³

Practical Application

Our results exemplify the importance of a well-designed regression model when estimating cost outcomes because simpler pre–post or cross-sectional comparisons without covariate adjustment can generate substantially different results. Still, we also recognize that payers cover actual (ie, unadjusted) expenses and adjustments for case mix and longitudinal trends do not directly affect a payer’s actual financial responsibility. Furthermore, most payers also have separate pharmacy and medical budgets. If a portion of the device or associated supply costs are covered by the pharmacy benefit, there would be a disconnect with the potential medical cost offsets associated with fewer diabetes complications.

An interesting policy question relates to the identification of appropriate population(s), if any, for new technologies. In the United States, CGMs and pumps are generally reimbursed for most T1 PwDs, although restrictions exist in Medicare.²⁴ We observed no meaningful differences between T1D and T2D for those above age 60, suggesting that the economic burden and benefits from technology are similar for T1D and T2D in the older age group. Although our study was not intended to identify the appropriate population, the results indicate that some factors, such as age, could be important considerations.

Provider capacity is another practical consideration. With both CGMs and pumps, we observed an increase in office visits, such as for endocrinologists. Reducing acute events is clearly crucial, but the number of endocrinologists, particularly those specializing in diabetes, has not kept up with the growth in the incidence of diabetes.²⁵ Nonspecialist providers likely will need to be involved to properly manage PwDs using technology. Furthermore, the COVID-19 pandemic has spurred much interest in telehealth and remote monitoring.²⁶ Diabetes technology facilitates virtual access to real-time information about insulin dosing and glucose levels, but provider concerns do exist, such as overwhelming data feeds and legal liabilities.²⁷

Limitations

First, we included extensive clinical and demographic covariates in the outcome models to adjust for the differences between the CGM and/or pump cohorts and our control, but it is possible that unmeasured confounders influenced the results. For example, the risk of hypoglycemia or hyperglycemia is presumably associated with CGM and/or pump use and the available covariates might not have adequately controlled for these risks or other factors associated with both device use and medical costs. It is also unknown whether our results generalize to other populations such as Medicaid beneficiaries or children. Second, the follow-up period was limited to one year, but pumps can have useful lives of five years or more. Furthermore, the device index date relied on a six-month clean period as sufficient to demonstrate device initiation. Some device users, namely those who buy supplies well ahead of need, likely were prevalent, rather than incident. Third, newer technology, such as hybrid closed loops and intermittent-scanning glucose monitors, were not widely available during the study period, so it is unknown whether our results are applicable to these devices. Fourth, hospitalization rates were decreasing for PwDs overall in the years leading up to the study time period,²⁸ so we might not have fully isolated the effects of the CGM and/or pump from background trends. Fifth, we did not consider pharmacy costs because of the lack of accurate insulin dosing information and the substantial manufacturer rebates for insulins. Similarly, while we excluded device costs from our analyses, any manufacturer rebates for devices would need to be considered to assess the overall economics of device use.

Future Work

Future work should examine longer follow-up periods, to help clarify the trajectory of relevant costs and benefits. In addition, newer devices need to be studied to determine whether updated technology yields different results. Identifying populations with the most favorable cost-benefit profile also remains a priority issue.

Conclusion

During a one-year follow-up, unadjusted medical costs of both CGM and pump appear lower than BGM, but multivariable modeling yielded adjusted savings only for CGM use. Economic benefits might be observable sooner for CGMs than for pumps. The GLMs fitted to health care utilization event rates, however, produced favorable results for both CGM and pump.

Supplemental Material

sj-xlsx-1-dst-10.1177_0042098020956930 – Supplemental material for Costs and Outcomes Comparison of Diabetes Technology Usage Among People With Type 1 or 2 Diabetes Using Rapid-Acting Insulin

Supplemental material, sj-xlsx-1-dst-10.1177_0042098020956930 for Costs and Outcomes Comparison of Diabetes Technology Usage Among People With Type 1 or 2 Diabetes Using Rapid-Acting Insulin by Carlos R. Vallarino, Siew H. Wong-Jacobson, Brian D. Benneyworth and Eric S. Meadows in Journal of Diabetes Science and Technology

Supplemental Material

sj-xlsx-2-dst-10.1177_0042098020956930 – Supplemental material for Costs and Outcomes Comparison of Diabetes Technology Usage Among People With Type 1 or 2 Diabetes Using Rapid-Acting Insulin

Supplemental material, sj-xlsx-2-dst-10.1177_0042098020956930 for Costs and Outcomes Comparison of Diabetes Technology Usage Among People With Type 1 or 2 Diabetes Using Rapid-Acting Insulin by Carlos R. Vallarino, Siew H. Wong-Jacobson, Brian D. Benneyworth and Eric S. Meadows in Journal of Diabetes Science and Technology

Footnotes

Acknowledgements

Both Caryl J. Antalis, PhD, and Nisha Narayanan, MPharm, from Eli Lilly and Company, assisted with the writing and editing.

Abbreviations

BGM, blood glucose monitor; CGM, continuous glucose monitor; CSII, continuous subcutaneous insulin infusion; DM, diabetes mellitus; ED, emergency department; GLM, generalized linear model; HbA1c, hemoglobin A1c; IPT, inverse probability of treatment; PwD, person with diabetes; QoL, quality of life; T1D, type 1 diabetes; T2D, type 2 diabetes.

Declaration of Conflicting Interests

The author(s) declared the following potential conflicts of interest with respect to the research, authorship, and/or publication of this article: C.R.V., S.H.W.J., and B.D.B. are employees and minor shareholders of Eli Lilly and Company. E.S.M. is a former employee of Eli Lilly and Company, currently employed by CIOX Health.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by Eli Lilly and Company.

ORCID iD

Carlos R. Vallarino

Supplemental Material

Supplemental material for this article is available online.

References

US Department of Health and Human Services, Centers for Disease Control and Prevention. National diabetes statistics report 2020. https://www.cdc.gov/diabetes/pdfs/data/statistics/national-diabetes-statistics-report.pdf. Accessed April 20, 2021.

American Diabetes Association. Economic costs of diabetes in the US in 2017. Diabetes Care. 2018;41(5):917-928.

Lynch

Riedel

Samant

, et al. Resource utilization with insulin pump therapy for type 2 diabetes mellitus. Am J Manag Care. 2010;16(12):892-896.

Cziraky

Abbott

Nguyen

, et al. A pragmatic clinical trial to compare the real-world effectiveness of V-Go versus standard delivery of insulin in patients with advanced type 2 diabetes. J Health Econ Outcomes Res. 2019;6(2):70-83.

Gill

Zhu

Shah

Chhabra

Health care costs, hospital admissions, and glycemic control using a standalone, real-time, continuous glucose monitoring system in commercially insured patients with type 1 diabetes. J Diabetes Sci Technol. 2018;12(4):800-807.

Fonda

Graham

Munakata

Powers

Price

Vigersky

RA.

The cost-effectiveness of real-time continuous glucose monitoring (RT-CGM) in type 2 diabetes. J Diabetes Sci Technol. 2016;10(4):898-904.

Quan

Sundararajan

Halfon

, et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Medical Care. 2005;43(11):1130-1139.

Quan

Couris

, et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am J Epidemiol. 2011;173(6):676-682.

U.S. Bureau of Labor Statistics. Measuring price change in the CPI: medical care. https://www.bls.gov/cpi/factsheets/medical-care.htm. Accessed May 12, 2021.

10.

Austin

Ghali

JV.

A comparison of several regression models for analysing cost of CABG surgery. Stat Med. 2003;22(17):2799-2815.

11.

Holm

A simple sequentially rejective multiple test procedure. Scand J Stat. 1979;6(2):65-70.

12.

Robins

Hernán

Brumback

Marginal structural models and causal inference in epidemiology. Epidemiology. 2000;11(5):550-560.

13.

Mihaylova

Briggs

O’Hagan

Thompson

SG.

Review of statistical methods for analysing healthcare resources and costs. Health Econ. 2011;20(8):897-916.

14.

Parkin

Graham

Smolskis

Continuous glucose monitoring use in type 1 diabetes: longitudinal analysis demonstrates meaningful improvements in HbA1c and reductions in health care utilization. J Diabetes Sci Technol. 2017;11(3):522-528.

15.

Bergenstal

Kerr

MSD

Roberts

Souto

Nabutovsky

Hirsch

IB.

Flash CGM is associated with reduced diabetes events and hospitalizations in insulin-treated type 2 diabetes. J Endocr Soc. 2021;5(4):bvab013.

16.

Sierra

Shah

Gill

, et al. Clinical and economic benefits of professional CGM among people with type 2 diabetes in the United States: analysis of claims and lab data. J Med Econ. 2018;21(3):225-230.

17.

Wan

Skandari

Minc

, et al. Cost-effectiveness of continuous glucose monitoring for adults with type 1 diabetes compared with self-monitoring of blood glucose: the DIAMOND randomized trial. Diabetes Care. 2018;41(6):1227-1234.

18.

Charleer

Mathieu

Nobels

, et al. Effect of continuous glucose monitoring on glycemic control, acute admissions, and quality of life: a real-world study. J Clinical Endocrinol Metab. 2018;103(3):1224-1232.

19.

Ackermann

Wallia

Kang

, et al. Comparative effectiveness and costs of insulin pump therapy for diabetes. Am J Manag Care. 2017;23(6):353-359.

20.

Grip

Svensson

Miftaraj

, et al. Real-world costs of continuous insulin pump therapy and multiple daily injections for type 1 diabetes: a population-based and propensity-matched cohort from the Swedish National Diabetes Register. Diabetes Care. 2019;42(4):545-552.

21.

Heller

White

Lee

, et al. A cluster randomised trial, cost-effectiveness analysis and psychosocial evaluation of insulin pump therapy compared with multiple injections during flexible intensive insulin therapy for type 1 diabetes: the REPOSE Trial. Health Technology Assess. 2017;21(20):1-278.

22.

Wan

Skandari

Minc

, et al. Cost-effectiveness of initiating an insulin pump in T1D adults using continuous glucose monitoring compared with multiple daily insulin injections: the DIAMOND randomized trial. Med Decis Making. 2018;38(8):942-953.

23.

Pease

Zomer

Liew

Earnest

Zoungas

Cost-effectiveness of health technologies in adults with type 1 diabetes: a systematic review and narrative synthesis. Syst Rev. 2020;9(1):171.

24.

Argento

Liu

Hughes

McAuliffe-Fogarty

AH.

Impact of Medicare continuous subcutaneous insulin infusion policies in patients with type 1 diabetes. J Diabetes Sci Tech. 2020;14(2):257-261.

25.

Vigersky

Fish

Hogan

, et al. The clinical endocrinology workforce: current status and future projections of supply and demand. J Clin Endocrinol Metab. 2014;99(9):3112-3121.

26.

Garg

Norman

GJ.

Impact of COVID-19 on health economics and technology of diabetes care: use cases of real-time continuous glucose monitoring to transform health care during a global pandemic. Diabetes Technol Ther. 2021;23(S1):S15-S20.

27.

Petrie

Peters

Bergenstal

Holl

Fleming

Heinemann

Improving the clinical value and utility of CGM systems: issues and recommendations: a joint statement of the European Association for the Study of Diabetes and the American Diabetes Association Diabetes Technology Working Group. Diabetes Care. 2017;40(12):1614-1621.

28.

Burrows

Gregg

Geiss

LS.

Declining rates of hospitalization for selected cardiovascular disease conditions among adults aged ≥35 years with diagnosed diabetes, US, 1998–2014. Diabetes Care. 2018;41(2):293-302.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.16 MB

0.01 MB