Nationally Subsidized Continuous Glucose Monitoring: A Cost-effectiveness Analysis

Type 1 diabetes presents challenges for optimal management, especially among children and adolescents. Long-term elevation of blood glucose levels is associated with premature vascular complications, while the debilitating impact of low blood glucose levels also reduces quality of life and productivity (1–6). Furthermore, only a minority of young people with type 1 diabetes are meeting glycemic targets, with international population data also showing that average glycemia is highest during adolescence (7–9). In this context, diabetes management technologies are hoped to reduce the burden of type 1 diabetes and improve outcomes.

While the use of continuous glucose monitoring (CGM) as well as insulin pump therapy is increasing internationally, cost is a barrier to equitable access (7,10,11). Prior to 2017, people with type 1 diabetes in Australia purchased CGM directly from manufacturers without any government subsidy or coverage by any private health insurer (12). Annual costs based on retail prices could exceed AUD 5,000 (USD 3,709 [2021]) depending on the brand selected, frequency of use, and promotional offers by manufacturers. In this context, CGM was previously used by only ∼5% of young people with type 1 diabetes in Australia (13). However, in April 2017 the Australian government introduced the national CGM Initiative, which provides universal access to CGM through completely subsidized funding of the technology for young people with type 1 diabetes (<21 years of age and some other groups) (12). Following the commencement of the CGM Initiative, use of CGM has been reported to have increased to ∼62–79% of young people nationally; however, the economic impact is unclear (10,13). In addition, insulin pumps are not routinely subsidized by the Australian government and access to insulin pump therapy is usually limited to those who can afford to purchase otherwise “optional” private health insurance with hospital cover (12). Despite a lack of subsidized access to insulin pumps, the Australian government does universally subsidize ∼90% or more of the cost of insulin, insulin pens, and insulin pump consumables (infusion sets and pump reservoirs/cartridges) (12). Any impact of subsidizing CGM on the uptake of insulin pump therapy, such as reducing the barrier of out-of-pocket cost for patients to access hybrid closed-loop therapy, could therefore influence overall population-level diabetes-management costs for the government.

As countries grapple with the increase in noncommunicable diseases (14), ensuring equitable access to efficacious and affordable management systems is a global priority and Sustainable Development Goal (15,16). Identifying the most effective and efficient funding models for diabetes-management systems including CGM is thus of international importance. Using longitudinal real-world data from a national Australian population-based study (13), we aimed to evaluate the cost-effectiveness of providing young people with access to completely subsidized CGM as well as to project the cost-effectiveness of extending the CGM Initiative to continue throughout adulthood for people with type 1 diabetes in Australia.

Development and reporting of our model followed the 2022 Consolidated Health Economic Evaluation Reporting Standards (CHEERS) statement (Supplementary Material) (17,18). A patient-level Markov model with annual cycles was used to evaluate the cost-effectiveness of access to universal CGM funding by the government in comparison to the previous system with completely user-funded CGM, among young people with type 1 diabetes aged 12 years, with follow-up until 21 years of age. This age range was chosen to mirror the population-based study that informed our economic analysis as well as existing age-based eligibility criteria for accessing subsidized CGM being largely restricted to those <21 years of age in Australia (13,19). A secondary analysis was performed to evaluate the cost-effectiveness of extending eligibility criteria for the CGM Initiative to everyone with type 1 diabetes over 21 years of age compared to a model of care with user-funded CGM. The perspective was that of the Australian health care system, and future costs and outcomes were discounted by 5% annually based on guidelines by the Australian Government Department of Health (20). The primary outcome was the incremental cost-effectiveness ratio (ICER) in terms of cost per quality-adjusted life-year (QALY) gained. The current Markov model represents an updated version of our recently reported author-created model (21), with changes outlined in Supplementary Material (pages 28–29).

Modeled health states comprised: type 1 diabetes with no complications or else any permutation of diabetic eye disease, diabetic nephropathy, diabetic foot disease, acute myocardial infarction (AMI), stroke, congestive heart failure (CHF), unstable angina, or death. Modeled stages of diabetic eye disease, diabetic nephropathy, and diabetic foot disease are outlined in Supplementary Tables 1 and 2. The regression of eye disease or renal disease was not directly modeled. The interrelated nature of vascular complications was reflected in using the shared risk factor of HbA_1c for health state transition probabilities as well as incorporating renal health states into annual cardiovascular risk prediction. Tracker variables were implemented to facilitate the independent modeling of health states representing different organ systems with their own disease progression. Mortality was modeled to be the result of cardiovascular disease, renal causes, or “other” background causes. Simulated individuals continued to cycle through the model from 12 up to 21 years of age in the primary analysis and up to death in the secondary analysis. Supplementary Fig. 1 presents the model structure.

Disease progression was simulated with transition probabilities derived from published sources (Supplementary Table 1). Contemporary data were sought through scoping reviews of the published literature as well as publicly accessible Australian databases for each modeled health state. Age-specific all-cause mortality was derived from published Australian data regarding people with type 1 diabetes (22).

Rates of severe and nonsevere hypoglycemia per person-year were modeled to reflect multiple hypoglycemic events in a year rather than annual transition probabilities per event. Hypoglycemic events were also assumed to be nonfatal. For prediction of the annual probability of a cardiovascular event, existing type 1 diabetes–specific cardiovascular risk prediction models were used and included HbA_1c and other important biomarkers as inputs (23,24) (Supplementary Table 3). Our Markov model used random numbers to determine patient sex, smoking status, and participation in regular exercise according to cohort-level probabilities as inputs within the cardiovascular risk prediction models. Recurrent cardiovascular events had to be preceded by any cardiovascular event at least 1 year earlier. Results from the Diabetes Control and Complications Trial (DCCT)/Epidemiology of Diabetes Interventions and Complications (EDIC) study were used to derive the proportion of events representing AMI, stroke, or CHF (25). Mortality rates for AMI, stroke, and CHF were subsequently derived from annualized probability and case-fatality data. Unstable angina and lower-extremity amputations were assumed to be nonfatal. Renal death could only occur in the presence of renal disease arising at least 1 year prior. Because projected cardiovascular and renal mortality rates were modeled as distinct health states impacted by glycemic control, previously reported estimates for these causes of death were subtracted from Australian all-cause mortality data to avoid double counting.

The modeled intervention was usual care following the introduction of subsidized access to CGM for young people with type 1 diabetes from 12 years of age, and follow-up until 21 years of age. The intervention incorporated any changes to use rates for CGM and insulin pump therapy, since the mode of glucose monitoring and insulin delivery may have impacted the observed glycemic outcomes, and the related consumables for insulin pumps are also subsidized by the Australian government. Based on the Australian population-based study with analysis of linked data from the Australasian Diabetes Data Network (ADDN) and the National Diabetes Services Scheme (NDSS), use and glycemic outcome data for those using CGM ≥75% and <75% of the time in the intervention arm were modeled separately (13). The modeled control group received usual care prior to subsidized access to CGM as defined by reported use rates for user-funded CGM or else capillary glucose testing in combination with multiple daily injections or insulin pump therapy. For the primary analysis, use rates of insulin pump therapy and CGM were derived from the same longitudinal Australian population-based study among those <21 years of age in the intervention and control arms (13). In the secondary analysis, the same use rates of CGM among those <21 years of age were applied over the life span, although use rates for insulin pump therapy among adults were taken from the Australian National Diabetes Audit (ANDA) (10). Use of capillary glucose testing or insulin injections was assumed to be the reciprocal of derived proportions for CGM or insulin pump therapy, respectively (Supplementary Tables 4 and 5).

All costs were reported in 2021 AUD, unless otherwise stated, with adjustment for inflation using the Consumer Price Index (health) (26). For the health state of type 1 diabetes without any complications, costs were sourced from Australia’s largest cost of illness study (27). We estimated costs for other health states using multiple sources (Supplementary Table 6).

Regarding intervention costs, the analysis included incremental costs of CGM when universally subsidized by the Australian government in comparison with an entirely user-funded model. Incremental costs also included any changes to the use of insulin pump therapy, assuming ongoing levels of government subsidy for insulin, insulin pens, and insulin pump consumables. Among those using CGM, it was assumed that each of the three currently available brands (Abbott Diabetes Care, AMSL Diabetes, and Medtronic Diabetes Australia) would be used in equal proportions. Among those using insulin pump therapy, an average cost for consumables was derived from available brands (Supplementary Table 6). We derived overall weighted costs for management from cohort-level use rates of insulin pumps, insulin injections, capillary glucose testing, and CGM among those using CGM ≥75% or <75% of the time according to age-group (<21 or ≥21 years of age). The previously described Australian cost of illness study presented aggregated costs of type 1 diabetes that captured the cost of insulin with multiple daily injections and capillary glucose testing (27). Consistent with current funding systems in Australia, the cost of insulin pump consumables but not insulin pumps was added to the cost of type 1 diabetes among those using insulin pump therapy.

The utility value for the health state of type 1 diabetes without complications was drawn from the largest Australian cost of illness study (27). Disutilities for the different stages of diabetic eye disease were derived from Japanese and Canadian patient preference-based time trade-off analyses that informed prior cost-effectiveness analyses (28–30). European analyses of people with type 2 diabetes or nephropathy or on chronic dialysis were implemented (31–33). For diabetic foot disease, European time trade-off analyses were selected that had been used previously in Australian cost-effectiveness analyses (34,35). A population-based time trade-off analysis in the United Kingdom among people with type 2 diabetes was used to derive disutilities for AMI, stroke, CHF, and angina. Annual disutilities of −0.0475 and −0.0041 were used for severe and nonsevere hypoglycemia, respectively. Disutility values for hypoglycemia among people with type 1 diabetes were derived from a large Canadian time trade-off survey (36). Disutilities were summed when multiple complications accrued. Supplementary Table 7 presents the utilities/disutilities used in analysis.

The model was populated by 10,000 simulated individuals profiled largely on the Australian linked data set from the ADDN and NDSS (13). Individuals entered the model at 12 years of age, with diabetes duration of 3 years and no complications. Rates of severe and nonsevere hypoglycemia across the life span were drawn from the Australian population-based study as well as large registry and cohort studies (2,13,37–44). The treatment effect of the CGM Initiative on HbA_1c determined the difference in transition probabilities between treatment groups for all vascular complications. Reductions in HbA_1c and rates of severe hypoglycemia among those with subsidized access to CGM compared with the previous user-funded model of care were drawn from the same Australian population-based study (13). The treatment effect of subsidized access to CGM on nonsevere hypoglycemia was derived from international pediatric trials and was compared with adult data to ensure consistency across the modeled time horizons (45–50). For the secondary analysis that included follow-up of simulated individuals from 12 years of age until death, clinical profiles and use rates of insulin pump therapy were supplemented with data from ANDA (10). A detailed profile of the simulated individuals and treatment effects is provided in Table 1.

Probabilistic sensitivity analyses were performed with 10,000 iterations to account for uncertainty across input parameters. Univariate scenario analyses also considered model assumptions that comprised reducing discount rates for future costs and outcomes to 3.5%, 1.5%, and 0%; assuming the baseline rate of severe hypoglycemia was 16.6 or 6.7 per 100 person-years based on local population studies (42,51); reducing the treatment effect of subsidized access to CGM on severe hypoglycemia by 50% and 100%; assuming no treatment effect of subsidized access to CGM on HbA_1c; increasing the baseline rate of nonsevere hypoglycemia to 104 events per person-year; reducing the treatment effect of subsidized access to CGM on nonsevere hypoglycemia by 25%; assuming the baseline HbA_1c was 1% (10.9 mmol/mol) higher or lower; assuming the average cost of CGM was equivalent to the highest (AUD 5,615) or lowest (AUD 2,412) recommended retail price of available CGM systems in Australia; reducing the baseline age of the modeled individuals to the youngest clinically recommended age for implementing CGM (i.e., 2 years of age with an assumed diabetes duration of 1 year); extending the model time horizon from 9 years to a lifetime; and over a lifetime horizon, assuming the treatment effect on HbA_1c was increased from −0.5% (5.5 mmol/mol) to −0.8% (8.7 mmol/mol). All univariate scenario analyses were replicated over a lifetime horizon for the secondary analysis, except for assuming reduced base rates of severe hypoglycemia on the basis of results reported among a pediatric population. Microsoft Excel (Microsoft, Redmond, WA) and TreeAge Pro Healthcare 2021 R1.1 (TreeAge Software LLC, Williamstown, MA) were used to complete the economic analyses.

Relative to an entirely user-funded model, government subsidy for the cost of CGM for young people up to 21 years of age was associated with gains of 0.46 (discounted) QALYs per person, at a net cost (discounted) of AUD 18,203. Over 10,000 iterations, the CGM Initiative was associated with an ICER of AUD 39,518 per QALY gained. At the commonly accepted willingness-to-pay (WTP) threshold of AUD 50,000 per QALY gained in Australia, 85% of simulations predicted the CGM Initiative to be cost-effective.

In scenario analyses we found that results were most sensitive to the cost of CGM as well as the treatment effect on nonsevere hypoglycemia and HbA_1c. Modeling the highest and lowest recommended retail price among the available brands of CGM in Australia changed the ICER to AUD 50,354 or AUD 21,125 per QALY gained, respectively. Assuming the base rate of nonsevere hypoglycemia was increased to 104 events per person-year led to an ICER of AUD 22,178 per QALY gained, while reducing the treatment effect on nonsevere hypoglycemia by 25% increased the ICER to AUD 52,359 per QALY gained. However, assuming no treatment effect of subsidized access to CGM on severe hypoglycemia only slightly increased the ICER to AUD 40,768 per QALY gained.

The secondary analysis that extended the time horizon to a lifetime decreased the ICER to AUD 34,890 per QALY gained. Over a lifetime horizon, assuming no treatment effect of subsidized access to CGM on HbA_1c increased the ICER to AUD 41,482 per QALY gained, while increasing the relative treatment effect from −0.5% (5.5 mmol/mol) to −0.8% (8.7 mmol/mol) reduced the ICER to AUD 33,994 per QALY gained. Similar to the primary analysis, results over a lifetime of follow-up were also sensitive to the cost of CGM as well as base rates of nonsevere hypoglycemia and treatment effects on nonsevere hypoglycemia from subsidized access to CGM. Results of the base-case and scenario analyses are presented in Tables 2 and 3, respectively.

The incremental cost-effectiveness plane and scatterplot (Fig. 1) present the variation in modeled outcomes for 10,000 iterations, with the CGM Initiative tending to provide greater effectiveness (QALYs) at an overall higher cost to government compared with a user-funded model of care. The cost-effectiveness acceptability curve over a lifetime (Supplementary Fig. 2) demonstrates the proportions of 10,000 iterations that were cost-effective for the CGM Initiative and a user-funded model of care at various WTP thresholds. Modeled clinical outcomes compared with empirical data, dependent external validation, and the Assessment of the Validation Status of Health-Economic decision models checklist are presented in supplementary material (Supplementary Tables 8–10 with supporting text as well as Supplementary Reporting and validation).

Our cost-effectiveness analysis produced an ICER of AUD 39,518 per QALY gained for the subsidized CGM Initiative relative to a user-funded model. Providing ongoing access to government-subsidized CGM is thus likely to be cost-effective in comparison to a system with user-funded CGM among young people with type 1 diabetes in Australia based on the commonly accepted WTP threshold of AUD 50,000 per QALY gained.

Our analyses of the CGM Initiative had a unique focus on young people from national real-world registries including longitudinal outcome and use data. Nonetheless, results were consistent with those of a systematic review that reported most international economic evaluations (62%) concluded CGM was cost-effective among predominantly adult populations (52). Previous analyses that reported CGM was cost-effective usually modeled an HbA_1c reduction ≥0.5% (5.5 mmol/mol) or else a more conservative HbA_1c reduction with a simultaneous reduction of hypoglycemia (52). This is consistent with efficacy results inputted into our model from our Australian national data set that reported those using CGM ≥75% of the time observed a 0.5–0.8% (5.5–8.7 mmol/mol) reduction in mean HbA_1c with an associated odds ratio for severe hypoglycemia of 0.59 (95% CI 0.48–0.78) (13). The secondary analysis considering a lifetime horizon was performed due to the incurable nature of type 1 diabetes. In the context of a lifetime horizon, the cost of CGM as well as base rates of nonsevere hypoglycemia and treatment effects on nonsevere hypoglycemia remained key drivers for cost-effectiveness. However, the relative impact of treatment effects on HbA_1c was greater in the secondary analysis likely because of its impact on vascular complications, which usually develop in adulthood. It is also notable that our economic evaluation showed that subsidized access to CGM was cost-effective in a “real-world” context by including attrition from the effective use of CGM as well as the potential impact on use of insulin pump therapy.

Strengths of the economic evaluation included an expert multidisciplinary team involved in model development and review for face validity. In addition, real-world data before and after the introduction of a national CGM funding initiative were implemented for modeled baseline glycemia, use rates of diabetes-management technologies, and treatment effects. This large data set spanning 3 years likely improves the generalizability of our economic evaluation for public health initiatives in comparison with analyses based on more selected populations involved in shorter randomized controlled trials. Modeling attrition from effective CGM use as well as the impact of funding CGM on other subsidized technologies was also expected to improve the generalizability of our results. Conservative estimates were used for the remaining model inputs with sensitivity analyses evaluating key drivers of cost-effectiveness. In addition, cost and utility data for the health state of type 1 diabetes without complications were sourced from Australia’s largest cost of illness study (27).

We also acknowledge some limitations of the study, including available data sources and underlying model assumptions. Due to the lack of consensus in the literature and the absence of relevant value sets for utility values in the pediatric population, it was assumed that utilities/disutilities were the same across all ages. In addition, the rate of nonsevere hypoglycemia events was not certain. The 2018 International Society for Pediatric and Adolescent Diabetes (ISPAD) guidelines acknowledged the likely underreporting of nonsevere episodes and estimated an average incidence of two events per person-week, or 104 events per person-year (44). Therefore, our assumption of baseline nonsevere hypoglycemia incidence being 55.7 events per person-year based on the UnderstandINg the impact of HYPOglycemia on Diabetes Management: A Survey of Perspectives and Practices (InHypo-DM) Study was considered conservative (38). Due to uncertainty regarding severe hypoglycemia incidence rates in the literature, we used data from the Australian population-based study in the base-case analysis and considered rates from other Australian studies in sensitivity analyses. Like most cost-effectiveness analyses, the projection of health outcomes over many years often relied on extrapolation from short-term clinical data. As with other validated economic models, utility values from populations with type 2 diabetes and from different countries were used when adequate detail was lacking regarding those with type 1 diabetes in Australia. Due to limitations of available data sources, the costs of end-stage kidney disease and of cardiovascular disease were based on general populations irrespective of comorbid diabetes and the cost of diabetic foot disease was based on a population with type 1 or type 2 diabetes. Results were also limited to the health care system perspective due to a lack of accessible data regarding the full social impact of type 1 diabetes in Australia. This was considered a conservative approach as any societal benefits from subsidized access to CGM would be expected to reduce the ICER further. Deterministic and probabilistic sensitivity analyses addressed uncertainty in the model. However, not all clinically relevant scenarios could be addressed through sensitivity analyses. In addition, the use of real-world data introduces other potential sources of bias. The NDSS implements the CGM Initiative, provides national access to all government-subsidized diabetes supplies, and is therefore assumed to have almost complete capture of individuals with type 1 diabetes in Australia. However, linked clinical outcome data through the ADDN were predominantly collected in large metropolitan centers. This approach may reduce generalizability of results for regional and remote communities and raises questions for future research regarding the equity of access to health services, despite universal funding for the CGM devices themselves.

Our results suggest that the recently implemented CGM Initiative would be cost-effective if extended beyond 21 years of age and support the extension of eligibility criteria for the government subsidy of CGM in Australia. However, in addition to cost-effectiveness analyses, further research is needed regarding budgetary impact, pregnancy-related outcomes, indirect costs, and optimal models of care for broader use that include improved clinical outcomes among those currently using CGM <75% of the time. Clinical trials and diabetes registries could collect standardized economic data to direct future economic evaluations and appraise implementation strategies. Furthermore, while most economic evaluations of diabetes-management technologies focus on the perspective of the health care system, more attention is needed on the equity of funding decisions and distributional effects as key factors in value-based health care, as well as the broader social and economic benefits that may be uncaptured by traditional health technology assessments.

In summary, we have demonstrated that government-subsidized access to CGM for young people with type 1 diabetes is likely to be cost-effective in comparison with a user-funded model in Australia. Providing subsidized access to CGM also appeared to represent good value for money when modeled eligibility was extended to adults for whom long-term HbA_1c reduction may have more pronounced clinical benefits through prevention of vascular complications to partially offset the long-term cost of technology.

Acknowledgments. The authors thank Peter van Wijngaarden (Centre for Eye Research Australia, Royal Victorian Eye and Ear Hospital, East Melbourne, Victoria, Australia, and Ophthalmology, Department of Surgery, The University of Melbourne, Parkville, Victoria, Australia) and Dr. Peter Heydon (Faculty of Medicine and Health, The University of Sydney, Camperdown, New South Wales, Australia; Royal Prince Alfred Hospital, Camperdown, New South Wales, Australia; and Liverpool Hospital, Liverpool, New South Wales, Australia) for expert advice regarding clinical assumptions in the model for diabetic eye disease as well as Jessica Klusty (Data for Decisions LLC, Waltham, MA) for critical review of the manuscript and supplementary information regarding model structure and description.

Funding. A.J.P. was supported by the Royal Australasian College of Physicians (RACP)/Diabetes Australia Research Establishment Fellowship from the RACP Foundation. This economic evaluation was developed through collaborations as part of the JDRF Global Centre of Excellence in diabetes research. Analysis of the linked ADDN and NDSS data sets was supported by JDRF Australia (4-SRA-2016-169-M-B), the recipient of the Australian Research Council Special Research Initiative in Type 1 Juvenile Diabetes. The NDSS is an initiative of the Australian government administered by Diabetes Australia.

The study funders had no role in study design; data collection, analysis, or interpretation; or manuscript preparation.

Duality of Interest. S.Z. reports participation on advisory boards or expert committees or in educational meetings outside the submitted work on behalf of Monash University for Boehringer Ingelheim, Eli Lilly, Sanofi, AstraZeneca, Novo Nordisk, and MSD Australia (payment to institution). Outside the submitted work, T.W.J. reports unrestricted support from Medtronic for investigator-led trials as well as grants from JDRF and Perth Children’s Hospital Foundation for study costs. D.J.H.-W. reports an educational grant from MSD Australia unrelated to the submitted work. D.E.B. reports a consulting contract with Edwards Lifesciences unrelated to the submitted work. E.A.D. reports participation in educational meetings on behalf of the Telethon Kids Institute for Eli Lilly (payment to institution). E.Z. has received grants from Amgen, AstraZeneca, Pfizer, and Shire outside the submitted work. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. A.J.P. performed analysis and drafted the manuscript. S.Z., E.C., T.W.J., S.R.J., D.J.H.-W., D.E.B., E.A.D., and E.Z. assisted with manuscript drafting and review. E.C. and E.Z. also provided advice regarding analysis. All authors read and approved the manuscript. S.Z. and E.Z. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the Australasian Diabetes Congress 2022 (Australian Diabetes Society/Australian Diabetes Educators Association Annual Scientific Meeting), Brisbane, Queensland, Australia, 8–10 August 2022.