Abstract
Rates of type 2 diabetes (T2D) continue to rise in the United States, with many patients failing to achieve glycemic targets. Primary care providers often serve as the sole clinician managing diabetes. Continuous glucose monitors (CGMs) have shown promise in diabetes management, yet their adoption in primary care settings, especially among patients with T2D not using intensive insulin therapy, remains limited. We sought to evaluate the impact of CGM use on glycemic control in patients with T2D not using insulin and those using basal but not bolus insulin in a primary care setting. CGM use was associated with a significantly greater reduction in HbA1c (-0.62%, p < 0.01) compared with matched controls at 3 months (n = 182). Patients showed improvements in time in range (39.7–61.9%, p < 0.0001), time > 180 mg/dL (60.1–37.9%, p < 0.001), time > 250 mg/dL (27.6–8.5%, p < 0.001), mean estimated glucose value (212 mg/dL to 173 mg/dL, p < 0.001) and glucose management indicator (8.39–7.46%, p < 0.001). CGM use in a primary care setting compared to usual care significantly improved glycemic control in T2D patients not on bolus insulin, irrespective of treatment with non-insulin or basal insulin. This suggests potential for broader CGM integration in primary care.
Supplementary Information
The online version contains supplementary material available at 10.1038/s41598-024-83548-4.
Keywords: Continuous glucose monitors (CGMs), Primary care, Type 2 diabetes, Insulin, Glycemic control, Noninsulin, Basal insulin
Subject terms: Endocrine system and metabolic diseases, Diabetes, Lifestyle modification, Preventive medicine, Endocrinology, Endocrine system and metabolic diseases
Introduction
Rates of type 2 diabetes continue to rise, and many patients with type 2 diabetes do not meet the recommended glycemic targets necessary to prevent complications associated with the disease1. For a large proportion of these patients, their primary care provider (PCP) serves as the sole clinician managing their diabetes2.
Continuous glucose monitors (CGMs) have emerged as a tool for diabetes management, with increasing use among patients with type 2 diabetes3,4. CGMs differ from self-monitoring blood glucose (SMBG) in that CGMs are worn passively and measure interstitial glucose values every 1–5 min over the 10- to 14-day sensor life, equating to 2,800 and 20,160 interstitial glucose readings per sensor or between 288 and 1,440 glucose readings per day5. CGMs allow patients to self-manage their diabetes by modifying their meal-related choices and allow for identification of glycemic patterns important for overcoming therapeutic inertia for medication-related changes6. Additionally, CGMs show directional trend arrows for interstitial glucose values; alert patients when glucose levels drop or are trending towards hypoglycemia or hyperglycemia allowing patients to identify and manage wide glycemic swings, and; identify episodes of nocturnal and treatment-induced hypoglycemia.
Randomized clinical trials (RCTs) and retrospective database studies have demonstrated an association between CGM utilization and reductions in HbA1c (A1c) as well as improvements in CGM metrics such as increased time in range (TIR, 70–180 mg/dL), decreased time spent in hypoglycemia (< 70 mg/dL) and reductions in the glucose management indicator (GMI) which estimates A1c using mean interstitial glucose data from the CGM7–10. CGM metrics are valuable to consider alongside A1c when assessing glycemic control since A1c is an indirect measure of average glycemia over 2–3 months with a highly variable correlation to mean glucose between patients, and is limited for the day-to-day management of diabetes11,12. Despite the known benefits of CGMs, adoption of CGMs in primary care for patients with type 2 diabetes has been limited, especially in patients not treated with insulin and in patients treated with basal but not bolus insulin13. In 2021, an analysis of 30,585 patients with type 2 diabetes showed CGM prescribing was more often associated with endocrinologists (78.6%) compared with PCPs (23.2%)14. Thus, increased adoption of CGMs by PCPs may also contribute to improve health disparities for patients with decreased access to a limited pool of endocrinologists. Disparities in specialty care occur in geographic regions with the highest diabetes prevalence and socioeconomic disadvantage; only 25% of United States (US) counties have an endocrinologist compared with 96% of counties with PCPs15,16.
A potential limiting factor for primary care prescribing of CGMs is that, while RCTs have demonstrated the efficacy of CGMs7,8, the effectiveness of CGM on glycemia in primary care patients with type 2 diabetes, particularly among those not on intensive insulin therapy, have not been well studied in real-world primary care settings.
The lack of evidence may have prevented some payers, such as Medicare, from covering CGMs for patients with type 2 diabetes who are not yet on insulin therapy17. Considering that approximately 80% of patients with type 2 diabetes in the US do not use insulin1,18, many patients who could potentially benefit from a CGM may have difficulty accessing or affording one.
In addition to payer coverage and reimbursement issues, recent studies have shown that PCPs have been hesitant to incorporate technology into their clinical practice19,20.
To address these gaps in the literature, we conducted a prospective, embedded pragmatic effectiveness study with retrospectively matched control patients that aimed to evaluate the impact of CGM use on glycemic control in patients with type 2 diabetes not using any insulin—and in patients using basal but not bolus insulin—in a primary care setting.
Methods
Study population
The study population included patients with a diagnosis of type 2 diabetes receiving primary care at a multispecialty medical group. Intervention patients (i.e., those who were provided with CGMs) were enrolled prospectively at six primary care clinics between October 2021 and August 2022. Informed consent was obtained.
Control patients were sourced from the medical group’s clinical EHR data through a previous partnership between AMGA (American Medical Group Association) and Optum®. Data were extracted, mapped, and normalized by Optum®.
Inclusion and exclusion criteria
To be included in the study, intervention patients must have been 18–85 years old at the time of enrollment, had a diagnosis of type 2 diabetes at least 90 days prior to enrollment, and had two documented readings of A1c ≥ 7.5. One A1c was taken on the day of enrollment or within 2 weeks prior. This became the patient’s baseline A1c. The second A1c must have been between 3 and 12 months prior to enrollment to confirm that the patient had a documented history of suboptimal A1c control.
To be included in the retrospective control pool, patients were required to have at least two primary care encounters at the study medical group that were at least three months apart between January 2018 and March 2020 or January and September 2021. The gap from March 2020 to January 2021 accounts for the disruption in care seen early during the COVID-19 pandemic. Like the intervention patients, control patients were required to be aged 18–85 at the time of the baseline A1c and must have had a diagnosis of type 2 diabetes at least 90 days prior to the baseline A1c. Additionally, control patients were required to have two recordings of A1c ≥ 7.5, plus a third A1c taken 3–4 months after the baseline A1c (i.e., the first recorded within study time frame). This became the control patients’ equivalent to the intervention patients’ 3-month A1c. This method established a pool of control patients that had two A1c readings ≥ 7.5, 3–12 months apart followed by a third A1c reading 3 months later, allowing for comparison of A1c difference over 3 months between intervention and control patients. Control patients were receiving usual care for their diabetes, compared to intervention patients who would be receiving usual care plus CGM.
Patients in both the intervention and control groups were excluded if they had gestational or chemically induced type 2 diabetes, end-stage renal disease (ESRD), were on hospice or receiving palliative care, were prescribed bolus or a combination of basal/bolus insulin in the past year, were pregnant, had alcohol or drug dependence, or ever had a documented history of personal CGM use. These exclusions were determined by the investigators to have potentially confounding effects on the outcome of change in A1c.
The number of basal insulin users that were permitted to enroll in the study was limited to ~ 25% of the total study population to ensure an adequate sample size of intervention patients who were not using insulin.
Study procedures
Before patients were enrolled, clinic staff viewed a webinar on how to implement the Dexcom™ G6 professional and personal CGM systems in their practice. This 90-minute webinar taught staff how to initiate patients on both the professional and personal CGM systems, obtain CGM reports through Dexcom CLARITY, and troubleshoot issues. The content of the webinar was based on information that is also readily available in the CGM user guide and on the manufacturer website. Most practice sites relied on medical office assistants (MOAs) to be trained to initiate participants on CGMs.
Patients were flagged based on the study inclusion and exclusion criteria as potentially being eligible for the study. Patients were offered the opportunity to participate in the study during a previously scheduled office visit. Once informed consent was obtained from the patients, they were oriented to the CGM system and smartphone application by clinic staff consisting of PCPs, physician assistants (PA), and MOAs. Patients used their own compatible smart device or were provided with one.
To establish baseline CGM metrics, the 3-month unblinded session (i.e., intervention period) was preceded by a 10-day CGM wear period with CGM data blinded to the patient and provider. The purpose of the blinded period was to establish baseline glucose data independent of patient behavioral modifications, which were expected to occur once patients were unblinded to their glucose measurements. After this 10-day wear period, patients switched to personal, i.e., unblinded CGMs for the remaining three months. Continuous glucose data from the personal CGMs were available to patients in real time via the smartphone application and to providers and clinic staff via a website. Patients were encouraged to monitor their glucose data and adjust their behavior (e.g., diet, exercise) to limit glycemic excursions. Providers were trained and encouraged to view patients’ glucose data and use it to inform discussions with the patients during the 3-month intervention (unblinded) period about controlling their glucose. All patients included in the primary analysis completed 3 months of personal CGM use.
After the initial study period of 3 months, all patients were offered the opportunity to extend their study participation and CGM use for an additional 3 months for a total of 6 months.
Measurement of glycemic control
Glycemic control was measured by comparing the difference between the A1c reading taken prior to CGM use with the A1c reading taken after 3 months of CGM use.
CGM data in the intervention patient group were assessed to calculate mean CGM metrics, which included percent time in range (TIR) (70–180 mg/dL), time below range (< 70 and < 54 mg/dL), time above range (> 180 and > 250 mg/dL), mean estimated glucose value or EGV (mg/dL) and GMI (%) for the 10-day blinded study period, the first 10 days of the unblinded 3-month study period (i.e., the unblinded baseline period), and the last 10 days of the unblinded 3-month study period (i.e., the unblinded 3-month period). Data were analyzed in 10-day intervals, not daily, e.g., 10-day mean EGV, because each individual CGM sensor lasts for 10 days. The percentage of patients who met TIR targets of 70% or more were also assessed for each of these study periods.
These definitions of glycemic control using CGM metrics align with American Diabetes Association (ADA) and American Association of Clinical Endocrinology (AACE) guidelines which state that, when CGM wear time is 70% or greater, a 10- to 14-day CGM assessment can be used to assess glycemic status and is useful in clinical management11,21.
Additionally, glycemic control was measured by calculating the proportion of patients in the intervention and control groups with A1c > 9 (poor control), A1c < 8 and A1c < 7 (good control) at baseline and 3 months and compared the difference-in-differences (DID). The A1c > 9 and A1c < 8 definitions align with the Healthcare Effectiveness Data and Information Set (HEDIS) measure A1c Control for Patients With Diabetes22 which has been revised and renamed as the Glycemic Status Assessment for Patients With Diabetes23 for 2024 and allows for the reporting of GMI from CGM use as an alternative to A1c. Per ADA guidelines, A1c < 7 is a definition of good control for glycemic targets for most adults with type 2 diabetes11.
Statistical analyses
All analyses were conducted in R version 4.2.2, and matching was conducted using the matchit package24,25. Intervention patients were matched 1:1 with patients from the control pool on baseline A1c within 0.2 points and age within 5 years.
Baseline characteristics of intervention and control patients were compared using two sample t-tests for continuous variables and chi square goodness of fit tests for categorical variables. Univariable and multivariable linear regression analyses were performed on the entire study population (i.e., all patients regardless of insulin usage) to assess the relationship between the change in A1c levels and intervention status, i.e., intervention (CGM) versus control (no CGM). A secondary set of regression analyses were performed on the population of patients who were not prescribed basal insulin.
Differences in mean CGM metrics between the blinded and first unblinded as well as the blinded and 3-month unblinded wear periods were evaluated using a Wilcoxon signed-rank test, two tailed with a Bonferroni-corrected significance level of p < 0.05. For the percentage of CGM users who met recommended TIR targets of 70% or more during the blinded baseline, the unblinded baseline, and unblinded 3-month periods a Chi-square test of independence was performed. To be included in the CGM analysis participants needed to have 70% of CGM data available during each study period.
Sub-analyses of the same CGM metrics were conducted among patients who extended their participation to 6 months. The difference in mean CGM metrics between the blinded and 6-month wear period was analyzed (in addition to the differences between the blinded baseline and unblinded baseline and the blinded baseline and unblinded 3-month study periods). As was done in the main analyses, CGM metrics were evaluated using a Wilcoxon signed-rank test, two tailed with a Bonferroni-corrected significance level of p < 0.05, except for TIR of 70% or more, where a Chi-square test of independence was performed. Patients were required to have 70% of CGM data available during all study periods to be included in the sub-analyses.
DID analyses were conducted using a linear regression with an interaction term between timepoint (baseline versus 3 months) and intervention status (control versus intervention) to determine if the change in the proportion of patients with A1c > 9, A1c < 8, and with A1c < 7 from baseline to 3 months was significantly different between the control and intervention groups.
This study received IRB approval from the Western Institutional Review Board (WIRB)-Copernicus Group, collectively known as the WCG Institutional Review Board. All experiments were performed in accordance with relevant guidelines and regulations.
Results
Baseline characteristics of the study population
A total of 182 patients were included in the analysis, which included 91 patients from the intervention and 91 matched patients from the control pool (Table 1). In the intervention arm, 24 patients were prescribed basal insulin compared to 41 in the control group, and 117 patients were not prescribed basal insulin (67 intervention patients, 50 control).
Table 1.
Baseline characteristics of the study population, control (no CGM) versus intervention (CGM) patients.
| Total | Control | Intervention | p value* | |
|---|---|---|---|---|
| (n = 182) | (n = 91) | (n = 91) | ||
| Baseline A1c % (Mean (SD)) | 9.2 (1.4) | 9.2 (1.3) | 9.2 (1.4) | 0.926 |
| Age Category (# (%)) | 1.00 | |||
| ≤44 | 12 (6.6) | 6 (6.6) | 6 (6.6) | |
| 45–54 | 47 (25.8) | 23 (25.3) | 24 (26.4) | |
| 55–64 | 59 (32.4) | 30 (33.0) | 29 (31.9) | |
| 65–74 | 56 (30.8) | 28 (30.8) | 28 (30.8) | |
| 75–85 | 8 (4.4) | 4 (4.4) | 4 (4.4) | |
| Sex (Female # (%)) | 80 (44.0) | 37 (40.7) | 43 (47.3) | 0.455 |
| Race (# (%)) | 0.641 | |||
| White | 131 (72.0) | 65 (71.4) | 66 (72.5) | |
| Black | 33 (18.1) | 19 (20.9) | 14 (15.4) | |
| Asian | 6 (3.3) | 2 (2.2) | 4 (4.4) | |
| American Indian or Alaska Native | 1 (0.5) | 0 (0.0) | 1 (1.1) | |
| Unknown | 11 (6.0) | 5 (5.5) | 6 (6.6) | |
| Insurance Type (# (%)) | 0.566 | |||
| Commercial | 106 (58.2) | 51 (56.0) | 55 (60.4) | |
| Medicare or Medicare Advantage | 66 (36.3) | 33 (36.3) | 33 (36.3) | |
| Medicaid | 4 (2.2) | 2 (2.2) | 2 (2.2) | |
| Self-pay | 5 (2.7) | 4 (4.4) | 1 (1.1) | |
| Other/Unknown | 1 (0.5) | 1 (1.1) | 0 (0.0) | |
| BMI Category (# (%)) | 0.884 | |||
| Normal (18.5–24.9) | 16 (8.8) | 8 (8.8) | 8 (8.8) | |
| Overweight (25.0–29.9) | 33 (18.1) | 14 (15.4) | 19 (20.9) | |
| Class 1 Obesity (30.0–34.9) | 53 (29.1) | 28 (30.8) | 25 (27.5) | |
| Class 2 Obesity (35.0–39.9) | 37 (20.3) | 18 (19.8) | 19 (20.9) | |
| Class 3 Obesity (40.0+) | 43 (23.6) | 23 (25.3) | 20 (22.0) | |
| ≥ 1 Dx † (# (%)) | ||||
| ASCVD§ | 44 (24.2) | 24 (26.4) | 20 (22.0) | 0.603 |
| Congestive Heart Failure (CHF) | 8 (4.4) | 6 (6.6) | 2 (2.2) | 0.278 |
|
Chronic Kidney Disease (CKD)/ Diabetic Kidney Disease (DKD) |
26 (14.3) | 19 (20.9) | 7 (7.7) | 0.02 |
| Hypertension | 149 (81.9) | 72 (79.1) | 77 (84.6) | 0.442 |
| Neuropathy | 30 (16.5) | 13 (14.3) | 17 (18.7) | 0.549 |
| Retinopathy | 9 (4.9) | 8 (8.8) | 1 (1.1) | 0.04 |
| ≥ 1 Rx † (# (%)) | < 0.001 | |||
| No OAD‡/GLP-1 | 1 (0.5) | 0 (0.0) | 1 (1.1) | |
| 1 OAD/GLP-1 | 17 (9.3) | 1 (1.1) | 16 (17.6) | |
| 2 + OAD/GLP-1 | 99 (54.4) | 49 (53.8) | 50 (54.9) | |
| Basal insulin +/- OAD/GLP-1 | 65 (35.7) | 41 (45.1) | 24 (26.4) |
*Bolded values are statistically significant at p < 0.05.
†Ascertained in the 12 months prior to baseline.
‡Oral antidiabetic, which includes biguanides, dipeptidyl peptidase-4 inhibitors (DPP4s), sodium-glucose cotransporter-2s (SGLT2s), sulfonylureas, and thiazolidinediones (TZDs).
§Atherosclerotic cardiovascular disease.
Patients in both groups had similar baseline characteristics; there were no significant differences between intervention and control patients on baseline A1c, age category, race, ethnicity, insurance type, body mass index (BMI), and presence of most chronic conditions except for stage 1–4 chronic kidney disease/diabetic kidney disease (CKD/DKD, as defined by presence of ICD-10 codes N18- for CKD or codes E10.2* or E11.2* for DKD), and retinopathy. CKD/DKD and retinopathy both had higher rates in the control group (Table 1). Diabetes prescription category at baseline was statistically significantly different between control and intervention patients (p < 0.001). Compared to control patients, fewer intervention patients were prescribed basal insulin, and more intervention patients were prescribed only 1 oral antidiabetic/glucagon-like peptide 1 (OAD/GLP-1). The median baseline A1c levels were comparable between the control and intervention groups.
Change in A1c: intervention vs. control
The average, unadjusted change in A1c for the intervention group was − 1.3% compared to -0.8% in the control group (Table 2). A large proportion of patients in both the intervention and control groups experienced declines in A1c, although the proportion that experienced a decline in A1c was larger in the intervention group (Table 3; Fig. 1).
Table 2.
Average unadjusted A1c values and change in A1c between baseline and 3 months in the control (no CGM) and intervention (CGM) groups.
| Total | Control | Intervention | p value* | |
|---|---|---|---|---|
| (n = 182) | (n = 91) | (n = 91) | ||
| Baseline A1c (Mean, (SD)) | 9.2 (1.4) | 9.2 (1.3) | 9.2 (1.4) | 0.93 |
| 3-month A1c (Mean, (SD)) | 8.2 (1.3) | 8.4 (1.4) | 7.9 (1.2) | < 0.01 |
| Change in A1c (Mean, (SD)) | -1.0 (1.5) | -0.8 (1.5) | -1.3 (1.5) | 0.01 |
*Two sample t-tests with bolded values statistically significant at p < 0.05.
Table 3.
Proportions of patients with good and poor A1c control at baseline and after 3 months in the control (no CGM) versus intervention (CGM) groups.
| Threshold | Control | Intervention | p value* |
|---|---|---|---|
| (n = 91) | (n = 91) | ||
| A1c < 7 | |||
| Baseline | 0% | 0% | |
| 3 months | 9% | 22% | |
| Difference | 9% | 22% | 0.01 |
| A1c < 8 | |||
| Baseline | 21% | 19% | |
| 3 months | 42% | 58% | |
| Difference | 21% | 40% | 0.04 |
| A1c > 9 | |||
| Baseline | 44% | 43% | |
| 3 months | 26% | 14% | |
| Difference | -18 | -29% | 0.24 |
*Results of difference-in-difference regression analysis controlling for timepoint, diabetes prescription regimen, and chronic kidney disease/ diabetic kidney disease with bolded values statistically significant at p < 0.05.
Fig. 1.
Difference (gray line) between baseline A1c (blue) and 3-month A1c (orange) for each patient in the control (left) and intervention (right) groups. After 3 months, intervention (CGM) patients had an average unadjusted change in A1c of -1.3% (SD = 1.5%, average baseline A1c = 9.2% (SD = 1.4%), average conclusion A1c = 7.9% (SD = 1.2%)) compared to -0.8% (SD = 1.5%, average baseline A1c = 9.2% (SD = 1.3%), average conclusion A1c = 8.4% (SD = 1.4%)) in the control group (p = 0.011).
The first multivariable model included all patients in the study, i.e., with or without a prescription for basal insulin (n = 181) and controlled for baseline diabetes medication regimen and CKD/DKD which were included in the model because the number of patients with these conditions were statistically significantly different between the control and intervention groups and because A1c may be less accurate for patients with CKD26. Use of CGM was associated with a 0.62% greater reduction in A1c levels compared with not using CGM (p < 0.01) (Table 4). The second multivariable model included only the subset of patients without a prescription for basal insulin (n = 117) and controlled for baseline diabetes medication regimen, diagnoses of CKD/DKD, and congestive heart failure (CHF), which was added to the model due to the statistically significant difference in number of patients with CHF between the control and intervention groups in this population subset. Use of CGM in this subset of patients was associated with a 0.66% greater reduction in A1c levels compared with not using CGM (p = 0.04).
Table 4.
Coefficients for linear regression model predicting the difference in A1c by intervention arm and by insulin prescription. CGM use was associated with a 0.62% greater reduction in A1c compared to the control (no CGM) group regardless of basal insulin prescription.
| Term | Estimate | Standard Error | t-value | p value |
|---|---|---|---|---|
| All patients with or without a prescription for basal insulin (n = 182) | ||||
| (Intercept) | -0.87 | 0.20 | -4.25 | < 0.01 |
| Ref (No CGM) | – | – | – | – |
| Intervention (CGM) | -0.62 | 0.24 | -2.57 | < 0.01 |
| Ref (2 + OAD/GLP-1) | – | – | – | – |
| No OAD/GLP-1 | 0.88 | 1.52 | 0.58 | 0.56 |
| 1 OAD/GLP-1 | 0.51 | 0.41 | 1.23 | 0.22 |
| Basal +/- OAD/GLP-1 | 0.27 | 0.24 | 1.10 | 0.27 |
| Ref (No CKD/DKD) | – | – | – | – |
| CKD/DKD | -0.02 | 0.33 | -0.07 | 0.95 |
| Patients without a prescription for basal insulin (n = 117) | ||||
| (Intercept) | -0.89 | 0.24 | -3.63 | < 0.01 |
| Ref (No CGM) | – | – | – | – |
| Intervention (CGM) | -0.66 | 0.31 | -2.11 | 0.04 |
| Ref (2 + OAD/GLP-1) | – | – | – | – |
| No OAD/GLP-1 | 0.95 | 1.54 | 0.62 | 0.54 |
| 1 OAD/GLP-1 | 0.57 | 0.42 | 1.35 | 0.18 |
| Ref (No CKD/DKD) | – | – | – | – |
| CKD/DKD | 0.53 | 0.45 | 1.19 | 0.24 |
| Ref (No CHF) | – | – | – | – |
| CHF | -0.62 | 0.72 | -0.86 | 0.39 |
| Patients with a prescription for basal insulin (n = 65) | ||||
| (Intercept) | -0.64 | 0.23 | -2.74 | 0.01 |
| Ref (No CGM) | – | – | – | – |
| Intervention (CGM) | -0.53 | 0.38 | -1.38 | 0.17 |
The multivariable DID analysis also controlled for baseline diabetes medication regimen and diagnoses of CKD/DKD with an added interaction term of timepoint by intervention status (Table 5). There was an increase in the proportion of patients with A1c < 7 and with A1c < 8 between the Baseline and 3-month time periods in both the intervention and control groups (Fig. 2). In the intervention group, there were 13.2% more patients with A1c < 7 (p = 0.013, Figs. 3) and 18.7% more patients with A1c < 8 (p = 0.046) compared with the controls. There were also 11% fewer patients with A1c > 9 in intervention group, but this difference was not statistically significant (p = 0.235, Fig. 4).
Table 5.
Coefficients for linear regression model predicting the difference in difference (DID) of the proportion of patients with good and poor A1c control from baseline to 3 months between intervention (CGM) and control (no CGM) groups.
| Term | Estimate | Standard Error | t-value | p value* |
|---|---|---|---|---|
| Proportion of patients with good control (A1c < 7) | ||||
| (Intercept) | 0.01 | 0.03 | 0.37 | 0.71 |
| Intervention Ref (No CGM) | – | – | – | – |
| Intervention (CGM) | 0.00 | 0.04 | -0.11 | 0.91 |
| Timepoint Ref (Baseline) | – | – | – | – |
| 3 Months | 0.09 | 0.04 | 2.34 | 0.02 |
| Intervention * Timepoint | 0.13 | 0.05 | 2.48 | 0.01 |
| Ref (2 + OAD/GLP-1) | – | – | – | – |
| No OAD/GLP-1 | -0.12 | 0.18 | -0.65 | 0.52 |
| 1 OAD/GLP-1 | 0.00 | 0.05 | 0.09 | 0.93 |
| Basal +/- OAD/GLP-1 | -0.02 | 0.03 | -0.81 | 0.42 |
| Ref (No CKD/DKD) | – | – | – | – |
| CKD/DKD | 0.00 | 0.04 | -0.08 | 0.93 |
| Proportion of patients with good control (A1c < 8) | ||||
| (Intercept) | 0.27 | 0.05 | 5.10 | < 0.01 |
| Intervention Ref (No CGM) | – | – | – | – |
| Intervention (CGM) | -0.05 | 0.07 | -0.80 | 0.42 |
| Timepoint Ref (Baseline) | – | – | – | – |
| 3 Months | 0.21 | 0.07 | 3.16 | < 0.01 |
| Intervention * Timepoint | 0.19 | 0.09 | 2.00 | 0.046 |
| Ref (2 + OAD/GLP-1) | – | – | – | – |
| No OAD/GLP-1 | 0.58 | 0.32 | 1.84 | 0.07 |
| 1 OAD/GLP-1 | 0.00 | 0.09 | -0.04 | 0.97 |
| Basal +/- OAD/GLP-1 | -0.15 | 0.05 | -2.91 | < 0.01 |
| Ref (No CKD/DKD) | – | – | – | – |
| CKD/DKD | 0.01 | 0.07 | 0.09 | 0.93 |
| Proportion of patients with poor control (A1c > 9) | ||||
| (Intercept) | 0.33 | 0.05 | 6.20 | < 0.01 |
| Intervention Ref (No CGM) | – | – | – | – |
| Intervention (CGM) | 0.03 | 0.07 | 0.46 | 0.65 |
| Timepoint Ref (Baseline) | – | – | – | – |
| 3 Months | -0.18 | 0.07 | -2.68 | 0.01 |
| Intervention * Timepoint | -0.11 | 0.09 | -1.19 | 0.24 |
| Ref (2 + OAD/GLP-1) | – | – | – | – |
| No OAD/GLP-1 | -0.22 | 0.31 | -0.70 | 0.49 |
| 1 OAD/GLP-1 | 0.04 | 0.08 | 0.52 | 0.60 |
| Basal +/- OAD/GLP-1 | 0.22 | 0.05 | 4.35 | < 0.01 |
| Ref (No CKD/DKD) | – | – | – | – |
| CKD/DKD | 0.05 | 0.07 | 0.70 | 0.48 |
*Bolded values are statistically significant at p < 0.05.
Fig. 2.
Change in the proportion of patients with good A1c control (< 7) from baseline to 3 months in the control (no CGM, green) versus intervention (CGM, red) groups (n = 182, p = 0.01*).
Fig. 3.
Change in the proportion of patients with good A1c control (< 8) Baseline to 3 months in the control (no CGM, green) versus intervention (CGM, red) groups (n = 182, p = 0.04*).
Fig. 4.
Change in the proportion of patients with poor A1c control (> 9) Baseline to 3 months in the control (no CGM, green) versus intervention (CGM, red) groups (n = 182, p = 0.24*). *Results of difference-in-difference regression analysis controlling for timepoint, diabetes prescription regimen, and chronic kidney disease/diabetic kidney disease with bolded values statistically significant at p < 0.05.
Change in CGM metrics: blinded, first unblinded and 3 month unblinded wear periods
For the 78 patients that met CGM data analysis inclusion criteria, there were significant improvements from blinded baseline to the unblinded baseline and unblinded 3-month study periods in percent TIR (70-180 mg/dL), percent time in hyperglycemia (> 180 and > 250 mg/dL), mean EGV and GMI (Table 6). There was little time spent in hypoglycemia during both the blinded and unblinded CGM wear periods.
Table 6.
CGM metrics (n = 78) for patients who met CGM analysis criteria. TIR (%), T > 180 (%), T > 250 (%), EGV (mg/dL), GMI (%), and % meeting target TIR improved after 3 months of unblinded CGM use compared to blinded CGM use at baseline.
| TIR (%) | T < 70 (%) | T < 54 (%) | T > 180 (%) | T > 250 (%) | EGV (mg/dL) | GMI (%) | Meets TIR Target (%) |
|
|---|---|---|---|---|---|---|---|---|
| Blinded baseline | 39.7 | 0.2 | 0.1 | 60.1 | 27.6 | 212 | 8.39 | 24 |
| Unblinded baseline | 61.9 | 0.2 | 0.0 | 37.9 | 8.5 | 173 | 7.46 | 46 |
| Difference (vs. blinded) | + 22.2 | 0.0 | 0.0 | -22.1 | -19.1 | -39 | -0.93 | + 22 |
| p value | < 0.001* | 0.42* | 0.33* | < 0.001* | < 0.001* | < 0.001* | < 0.001* | 0.007† |
| Unblinded 3-Month | 62.2 | 0.4 | 0.1 | 37.4 | 9.7 | 173 | 7.46 | 47 |
| Difference (vs. blinded) | + 22.5 | + 0.2 | 0.0 | -22.7 | -17.8 | -39 | -0.93 | + 23 |
| p value | < 0.001* | 0.08* | 0.77* | < 0.001* | < 0.001* | < 0.001* | < 0.001* | 0.005† |
Key: (TIR = time in range, T = time, EGV = estimated mean glucose, GMI = glucose management indicator). All glucose values reported as mg/dL. Bonferroni correction for 8 hypothesis tests with family-wise error rate of 0.05.
*Wilcoxon signed-rank test, two-tailed.
†Chi-square test of independence.
The unblinded baseline wear period showed a significant improvement from blinded in percent TIR (39.7–61.9%, p < 0.0001), percentage of time > 180 mg/dL (60.1–37.9%, p < 0.001), percentage of time > 250 mg/dL (27.6–8.5%, p < 0.001), EGV (212 mg/dL to 173 mg/dL, p < 0.001) and GMI (8.39–7.46%, p < 0.001). These improvements were all sustained during the 3-month unblinded wear period.
In addition, during the blinded baseline study period, 24% of patients met recommended TIR goals of 70% or more which increased significantly to 46% and 47% of patients meeting TIR goals during the unblinded baseline and unblinded 3-month study periods, respectively.
The observed changes in TIR from the blinded baseline to unblinded baseline periods of 39.7–61.9% (p < 0.0001) equates to 5.3 more hours a day spent in range, and the observed changes in percentage of time > 180 (60.1–37.9%) and > 250 mg/dL (27.6–8.5%, p < 0.001) equates to 5.3 and 4.6 fewer hours a day spent in hyperglycemia.
Change in CGM metrics: 6-month sub-analysis
For the 37 patients that met CGM data analysis inclusion criteria for the 6-month sub-analysis, significant improvements were observed from the blinded baseline to the unblinded 6-month wear period (in addition to the unblinded baseline and unblinded 3- month study periods) in percent TIR (70-180 mg/dL), percent time in hyperglycemia (> 180 and > 250 mg/dL), mean EGV and GMI (Table 7). There was little time spent in hypoglycemia during all CGM study periods.
Table 7.
CGM metrics (n = 37) for patients who met CGM analysis criteria after 6 months of CGM use. Improvements in TIR (%), T > 180 (%), T > 250 (%), EGV (mg/dL), and GMI (%) persisted after 6 months of CGM use.
| TIR (%) | T < 70 (%) | T < 54 (%) | T > 180 (%) | T > 250 (%) | EGV (mg/dL) | GMI (%) | Meets TIR Target (%) | |
|---|---|---|---|---|---|---|---|---|
| Blinded baseline | 41.9 | 0.1 | 0.0 | 58.0 | 23.1 | 205 | 8.2 | 27 |
| Unblinded baseline | 66.9 | 0.2 | 0.0 | 32.9 | 4.9 | 167 | 7.3 | 54 |
| Difference (vs. blinded) | + 25.0 | + 0.1 | 0.0 | -25.1 | -18.2 | -38 | -0.9 | + 27 |
| p value | < 0.001* | 0.06* | 1.00* | < 0.001* | < 0.001* | < 0.001* | < 0.001* | 0.03† |
| Unblinded 3-month | 68.9 | 0.5 | 0.1 | 30.6 | 5.7 | 165 | 7.3 | 60 |
| Difference (vs. blinded) | + 27.0 | + 0.4 | + 0.1 | -27.4 | -17.4 | -40 | -1.0 | + 32 |
| p value | < 0.001* | 0.13* | 0.27* | < 0.001* | < 0.001* | < 0.001* | < 0.001* | 0.01† |
| Unblinded 6-month | 65.5 | 0.2 | 0.1 | 34.3 | 10.1 | 172 | 7.4 | 54 |
| Difference (vs. blinded) | + 23.6 | + 0.1 | 0.0 | -23.7 | -13.0 | -33 | -0.8 | + 27 |
| p value | 0.001* | 0.55* | 0.29* | 0.001* | 0.002* | < 0.001* | < 0.001* | 0.03† |
Key: (TIR = time in range, T = time, EGV = estimated mean glucose, GMI = glucose management indicator). All glucose values reported as mg/dL. Bonferroni correction for 8 hypothesis tests with family-wise error rate of 0.05. Changes in green are significant, based on significance threshold of 0.006.
*Wilcoxon signed-rank test, two-tailed.
†Chi-square test of independence.
The unblinded 6-month wear period showed a significant improvement compared to the blinded baseline period in percent TIR (41.9–65.5%, p = 0.001), percent time > 180 mg/dL (58–34.3%, p = 0.001), percent time > 250 mg/dL (23.1–10.1%, p = 0.002), EGV (205 mg/dL to 172 mg/dL, p < 0.001) and GMI (8.2–7.4%, p < 0.001).
No adverse events related to the intervention were reported during this study.
Discussion
Overall, use of CGMs in a primary care setting significantly improved glycemic control in patients with type 2 diabetes who were not prescribed bolus insulin compared with matched control patients, irrespective of whether patients were treated with non-insulin or with basal insulin. The 0.62% greater reduction in A1c levels observed in CGM users compared to non-CGM users (p < 0.01) exceeds the generally accepted threshold of 0.5% for a clinically significant change in A1c27. The proportion of patients in this population achieving good A1c control (A1c < 7 and A1c < 8) was also significantly greater among those that used CGM.
The greater reduction in A1c for the CGM intervention group (p < 0.01) may be attributed to patient behavior modifications such as diet and/or lifestyle changes. This is further supported by significant improvements in TIR (70-180 mg/dL), EGV, GMI and other CGM metrics from the blinded baseline period to the first unblinded and 3-month periods. These improvements were sustained through 6 months among patients who elected to continue in the study. After the unblinding of CGM data during the first unblinded baseline period, patients could presumably make decisions to self-manage their glucose levels to increase their TIR and lower their EGV, GMI and time spent in hyperglycemia.
Overall, these results align with previous literature that demonstrated the effect of CGMs on A1c and CGM metrics among patients prescribed various regimens of insulin7–10, but our study demonstrates the potential for integration into primary care settings, particularly for patients on less intensive or no insulin therapy.
Strengths
Our study was conducted in a real-world primary care setting, which reflects conditions that are more generalizable and representative of settings in which many patients with type 2 diabetes receive diabetes care. Our study also utilized matched controls receiving care at the same organization, reducing the influence of secular trends unrelated to the intervention compared to a single arm pre- post-study design.
Limitations
Our study was limited because of its 3-month duration (which may not be sufficient to capture long-term effect), being conducted in one health system, a lack of temporal matching (i.e., the use of retrospective controls), and a lack of randomization. Control patients were not enrolled prospectively; they were derived from the EHR, so control patients were not able to be given CGMs. Provider medication changes based on A1c and/or CGM data may be more likely to occur at the 3-month mark during an in-person visit and a longer duration study assessing the impact of these changes on glycemic control would be of value. The 6-month sub-analysis was limited; only a subset of patients elected to continue study participation, which reduced the sample size. There is also potential selection bias among those patients electing to continue the study.
Although prescriptions for basal insulin and CKD/DKD diagnosis were controlled for in the regression model, there were significantly more patients in the control group that were prescribed basal insulin and had a diagnosis of CKD/DKD, suggesting that control patients were overall further along in their diabetes progression than the intervention patients.
Additionally, ethnicity information was unable to be captured from the electronic health record, and there was a small overall number of African Americans and other racial categories such as Asian. Also, only the CGM intervention group had CGM data to assess since the control group was retrospectively matched and did not wear CGM devices. CGM data was limited in that we were not able to more granularly assess the impact that daily events, such as meals or sleep, had on the overall 10-day CGM measurements.
Another limitation was that patients with ESRD, i.e., stage 5 CKD/DKD, were excluded from the study. CGM may also be beneficial for patients with ESRD since A1c levels are not as accurate for these patients28.
Future research should aim to include a more diverse patient population, add more healthcare organizations, and further extend the study period to better understand the long-term impact of CGM use. Since control patients were derived from the EHR and not enrolled prospectively, future research could also randomize patients to treatment (unblinded CGM) and control (blinded CGM) study arms temporally to observe differences in glycemic outcomes in patients who were blinded to CGM data for the full duration of the study compared to patients who were unblinded. Studies could also be conducted for patients with pre-diabetes and for patients with ESRD.
Implications
Our study adds to the evidence supporting the efficacy of CGMs in improving glycemic control in a real-world, primary care setting, particularly among patients with type 2 diabetes who are prescribed basal insulin without bolus insulin or are not prescribed any type of insulin, though further studies in this area are warranted. Ensuring that CGMs are accessible and prescribed to more patients treated in primary care may drive better outcomes for patients with type 2 diabetes regardless of medication treatment.
Supplementary Information
Below is the link to the electronic supplementary material.
Acknowledgements
The authors thank all clinicians, staff members and participants at the medical group who were involved in the study. We’d also like to thank Caitlin Shaw, Lauren Jepson and Megan Zachary for data assistance and Jeff Mohl, Meghana Tallam, and Carlos Moreno for analytical support.
Author contributions
S.S. assisted with protocol development and led study operations, data collection, data analysis and drafted the manuscript text. R.T. assisted with protocol development, served as a subject matter expert, and assisted with manuscript development. J.D., J.M., and J.C. assisted with protocol development and led patient recruitment, study and clinical operations, and data collection. E.C. was the principal investigator, led protocol development, and oversaw data analysis and manuscript development. All authors reviewed the manuscript.
Funding
Mr. Shields and Dr. Ciemins report grants from Dexcom, Inc. during the conduct of the study given to their institution of employment (AMGA). Dr. Thomas is an employee of Dexcom, Inc. Mrs. Durham, Dr. Moran, and Mr. Clary report grants from AMGA during the conduct of the study given to their institution (Piedmont HealthCare).
Data availability
The data generated and analyzed during the current study are not publicly available because data were obtained from a private source (Optum®) that does not allow public sharing of their data. Data are available from the corresponding author upon reasonable request.
Declarations
Competing interests
The authors declare no competing interests.
Footnotes
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
References
- 1.National Diabetes Statistics Report. Centers for Disease Control and Prevention.
- 2.Healy, A. M. et al. Endocrinologists’ opinions of diabetology as a primary care subspecialty. Clin. Diabetes. 36(2), 168–173 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 3.Kompala, T. & Neinstein, A. A new era: Increasing continuous glucose monitoring use in type 2 diabetes. Am. J. Manag Care25(4 Spec No.), Sp123–sp126 (2019). [PubMed] [Google Scholar]
- 4.Pathak, S. et al. Impact of expanding access to continuous glucose monitoring systems among insulin users with type 1 or type 2 diabetes. Diabetes Technol. Ther.25(3), 169–177 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Fico, G. et al. Exploring the frequency domain of continuous glucose monitoring signals to improve characterization of glucose variability and of diabetic profiles. J. Diabetes Sci. Technol.11(4), 773–779 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Miller, E. M. Using continuous glucose monitoring in clinical practice. Clin. Diabetes. 38(5), 429–438 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Cox, D. J. et al. Minimizing glucose excursions (GEM) with continuous glucose monitoring in type 2 diabetes: A randomized clinical trial. J. Endocr. Soc.4(11), bvaa118 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Martens, T. et al. Effect of continuous glucose monitoring on glycemic control in patients with type 2 diabetes treated with basal insulin: A randomized clinical trial. JAMA325(22), 2262–2272 (2021). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Shields, S. et al. HbA1c improvements after initiation of real-time continuous glucose monitoring in primary care patients with type 2 diabetes. J. Diabetes Sci. Technol.17(5), 1423–1424 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10.Beck, R. W. et al. Continuous glucose monitoring versus usual care in patients with type 2 diabetes receiving multiple daily insulin injections: A randomized trial. Ann. Intern. Med.167(6), 365–374 (2017). [DOI] [PubMed] [Google Scholar]
- 11.Standards of Care in Diabetes—2024. American Diabetes Association Professional Practice Committee. (2024).
- 12.Beck, R. W. et al. The fallacy of average: How using HbA(1c) alone to assess glycemic control can be misleading. Diabetes Care. 40(8), 994–999 (2017). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Martens, T. W. & Parkin, C. G. How use of continuous glucose monitoring can address therapeutic inertia in primary care. Postgrad. Med.134(6), 576–588 (2022). [DOI] [PubMed] [Google Scholar]
- 14.Mayberry, L. S. et al. Rates and correlates of uptake of continuous glucose monitors among adults with type 2 diabetes in primary care and endocrinology settings. J. Gen. Intern. Med.38(11), 2546–2552 (2023). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Lu, H. et al. Population-based geographic access to endocrinologists in the United States, 2012. BMC Health Serv. Res.15, 541 (2015). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Lòpez-DeFede, A. & Stewart, J. E. Diagnosed diabetes prevalence and risk factor rankings, by state, 2014–2016: A ring map visualization. Prev. Chronic Dis.16, E44 (2019). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.AARP. Does Medicare cover continuous glucose monitors? Medicare Q&A 2023 June 7, 2023 [cited 2023 9/28]; https://www.aarp.org/health/medicare-qa-tool/does-medicare-cover-continuous-glucose-monitors.html
- 18.Cefalu, W. T. et al. Insulin access and affordability working group: Conclusions and recommendations. Diabetes Care. 41(6), 1299–1311 (2018). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Grunberger, G. et al. Treatment intensification with insulin pumps and other technologies in patients with type 2 diabetes: Results of a physician survey in the United States. Clin. Diabetes. 38(1), 47–55 (2020). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Oser, T. K. et al. Continuous glucose monitoring in primary care: Understanding and supporting clinicians’ use to enhance diabetes care. Ann. Fam Med.20(6), 541–547 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Grunberger, G. et al. American association of clinical endocrinology clinical practice guideline: The use of advanced technology in the management of persons with diabetes mellitus. Endocr. Pract.27(6), 505–537 (2021). [DOI] [PubMed] [Google Scholar]
- 22.National Committee for Quality Assurance (NCQA). HEDIS measure: Hemoglobin A1c Control for Patients With Diabetes. https://www.ncqa.org/wp-content/uploads/2022/07/HEDIS-MY-2023-Measure-Description.pdf (2022).
- 23.HEDIS MY 2024 Measures and Descriptions. National Committee for Quality Assurance (NCQA).
- 24.Ho, D. et al. MatchIt: Nonparametric preprocessing for parametric causal inference. J. Stat. Softw.42(8), 1–28 (2011). [Google Scholar]
- 25.R Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, 2022). [Google Scholar]
- 26.Speeckaert, M. et al. Are there better alternatives than haemoglobin A1c to estimate glycaemic control in the chronic kidney disease population? Nephrol. Dialysis Transplantation. 29(12), 2167–2177 (2014). [DOI] [PubMed] [Google Scholar]
- 27.Little, R. R., Rohlfing, C. L. & Sacks, D. B. Status of hemoglobin A1c measurement and goals for improvement: From chaos to order for improving diabetes care. Clin. Chem.57(2), 205–214 (2011). [DOI] [PubMed] [Google Scholar]
- 28.Ling, J. et al. Use of continuous glucose monitoring in the assessment and management of patients with diabetes and chronic kidney disease. Front. Endocrinol. (Lausanne). 13, 869899 (2022). [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Data Availability Statement
The data generated and analyzed during the current study are not publicly available because data were obtained from a private source (Optum®) that does not allow public sharing of their data. Data are available from the corresponding author upon reasonable request.