Impact of Continuous Glucose Monitoring Initiation on Emergency Health Services Utilization

We assessed the effects of initiation of continuous glucose monitoring (CGM) on emergency department (ED) visits among individuals covered by a large commercial health insurer in the southeastern U.S. (1).

Our sample includes beneficiaries with diabetes who were using insulin between 1 December 2017 and 31 December 2020. To be included, member–index date combinations had to 1) have four or more adjudicated claims for insulin and one or more medical claims with a diabetes diagnostic code in the 365 days before index, 2) be eligible for ≥10 months throughout the preperiod, 3) not turn 65 years old before the end of the postperiod, and 4) have one or more health care charges in the pre- or postperiod.

We implemented a pre-post nonequivalent groups design via differences-in-differences (DiD) with administrative claims data, and we used generalized estimating equations with a negative binomial distribution, log link, and exchangeable correlation structure to model all-cause ED visits per 1,000 members.

The intervention date (among the treated individuals) was the first date of a filled prescription for any personal-use CGM (real time or intermittently scanned) after 1 December 2018, the policy implementation date. The index date for the control group was the day of a health care encounter that included both a diabetes-related diagnostic code and an evaluation and management code. The DiD timeline included 180 days of observation before and after the index date.

To reduce confounding, we used a two-phase, one-to-one matching approach across the treatment and control groups. We first exact-matched control individuals to each treatment individual on endocrinology office visits, claim-identifiable hypoglycemic events, insulin type, source of insurance, diabetes type, professional CGM use in the preindex period, and year and season of index. Subsequently, we used a propensity score model among this set that further accounted for age at index, chronic conditions (2), residential rurality, and sex. Data on race and ethnicity were not available for our analytic data set. Our final set of matched individuals was selected using the control individual who had the minimum absolute distance of propensity score from the treated individual.

Data from 6,180 individuals (3,090 in each group) were analyzed. Standardized absolute mean differences across all variables were <0.1 after matching. Roughly half saw an endocrinologist in the prior year, 17% used long-acting insulin only, 18% used rapid-acting insulin only, and 65% used both. Approximately one-fourth lived in a rural zip code, and slightly more than half had type 2 diabetes (versus type 1). The DiD estimate was an incidence rate ratio (IRR) of 0.86 (95% CI 0.74, 1.00), representing a 14% lower incidence rate of ED visits for CGM initiators compared with those who did not initiate CGM (rates are depicted in Fig. 1). When stratified, the effect estimate was an IRR of 0.83 for those with type 1 diabetes and 0.88 for those with type 2 diabetes, although the subgroup estimates were less precise and not statistically significant.

Our analysis, based on data from beneficiaries of a large private health insurer in the southeastern U.S., underscores the potential for CGM to reduce emergency health service use among people with diabetes who also use insulin. Limitations include that the data do not capture individuals using cash to pay for CGM, although this would conservatively bias results toward the null by reducing the pre-post change difference between groups. We were also unable to evaluate CGM initiation and outcomes across different racial and ethnic subgroups. Our inclusion criteria for person-time combinations make parallel trend testing infeasible, so the estimated treatment effect may be attributable to differing trends in ED use between the groups prior to CGM initiation (3). Our matching approach led us to estimate the treatment effect in the treated individuals; thus, findings are not generalizable to adults with diabetes who are not on CGM.

Statistically significant associations between CGM initiation and several emergency health service use measures, including ED visits for hypoglycemia, although all-cause ED visits were not statistically significant in this association (4). Our estimates are similar in magnitude and have greater precision. However, we could not replicate the outcomes evaluated in the aforementioned study due to the limited number of ED visits; more work is needed to evaluate the broad spectrum of CGM-derived health outcomes across diverse cohorts.

Growing empirical support for a link between CGM use and reduced emergency health services use is nontrivial in the context of the economic, clinical, and patient-oriented burden of diabetes and may provide evidence for health insurers that aim to minimize the frequency of ED visits. Additionally, based on previous research that showed notably more hypoglycemia episodes occur at home and do not result in an ED visit, CGM may affect hypoglycemia that is not captured in claims data (5).