Association Between Therapeutic Inertia and Future Hypoglycemia Among Patients with Type 2 Diabetes

Abstract

Introduction:

Therapeutic inertia (TI), the failure to adjust therapy when HbA1c remains above- target, is a barrier to optimal glycemic control in type 2 diabetes (T2D). This study examined the association between TI and subsequent-year hypoglycemia visit, and whether continuous glucose monitoring (CGM) modifies this relationship.

Methods:

We analyzed electronic health records (2017–2023) from two Midwest US healthcare systems, including adults with T2D, at least one above-target HbA1c (>7% for ages 18–64; >8% for ages ≥65), and glucose-lowering prescriptions. TI was calculated annually as the percentage of above-target HbA1c results without prescription changes within 30 days. We fitted logistic regression models to examine whether high TI (>50%) was associated with subsequent-year hypoglycemia visits. An interaction term tested whether this association differed between those who used vs. did not use CGM.

Results:

Among 65,983 participants (mean age 56, 51% male, 75% White), mean HbA1c at last follow-up was 8.1% (ages 18–64) and 8.0% (ages ≥ 65). High TI was associated with 31% increased odds of hypoglycemia visit (OR=1.74; 95% CI: 1.61–1.88; p<0.001). Insulin users had threefold higher odds (OR=2.95; p<0.001). Medicare beneficiaries had 31% higher odds than Medicaid beneficiaries. Adults aged 18–44 years had more hypoglycemia visits compared to other age groups. Low CGM use (7%) limited the interpretation of interaction effects (95% CI: 0.47–1.1, p=0.12).

Conclusions:

In this cohort, high TI predicted hypoglycemia-visit. Further research is needed to understand TI drivers and how to balance improving glycemic control without increasing the risk of hypoglycemia.

1. Introduction

Therapeutic inertia (TI) contributes to poor diabetes management. The American Diabetes Association (ADA) defines TI as the failure to make timely treatment adjustments when a patient’s hemoglobin A1c (HbA1c) remains not within target [1]. The median delay in treatment change exceeds one year, ranging from 0.3 to 7.2 years [2]. This delay compromises glycemic management and heightens the risk of serious complications such as kidney failure, amputation, and blindness [3]. TI also disproportionately affects high-risk or clinically complex patients, for whom clinicians are less likely to modify drug therapy even when indicated [4].

Patients receiving insulin therapy experience elevated rates of TI because dosage adjustments are highly complex and frequently associated with hypoglycemia (glucose level of <69 mg/dL) [5]. Despite clinical need, insulin intensification occurs in only 25–30% of eligible patients and is discontinued at comparable rates [6]. Advanced age further exacerbates therapeutic inertia due to the presence of multiple comorbidities, which complicates medication modifications. Clinician-related barriers include a hesitancy to intensify treatment, under-prescribing sulfonylureas and glinides, overestimating the patient’s ability to recognize hypoglycemia symptoms (such as shakiness, sweating, and confusion), and insufficient training in hypoglycemia management [7]. Clinicians also must consider drug adverse effects, drug interactions, and patient adherence to drug regimens [8]. Insurance barriers represent another significant contributor to therapeutic inertia, because Medicare Part D historically required substantial out-of-pocket costs (up to $1,000) for highly effective agents such as SGLT2 inhibitors and GLP-1 receptor agonists [9].

Continuous glucose monitoring (CGM) offers a promising approach to reduce TI in diabetes management. CGM reports provide an assessment of glycemic patterns and variability that is far more comprehensive than the view provided by HbA1c (the average blood glucose level over 3 months) [10]. Modern CGM systems do more than simply display current glucose levels; they offer predictive alerts for future hypoglycemic events and actively assist users in making informed decisions to prevent common adverse effects of hypoglycemia from using insulin and sulfonylureas [11]. Despite its promising benefits, CGM remains underutilized. CGM use is substantially lower among Black (28%) and Hispanic (37%) young adults than among their White (71%) peers [12], highlighting stark racial disparities. These disparities are compounded by inconsistent insurance coverage; for example, many Medicaid beneficiaries are denied access due to restrictive eligibility criteria and the lack of a consistent national policy [13].

A critical gap remains in understanding how insurance coverage, insulin use, and CGM availability interact to influence TI in real-world clinical settings. Although TI is detrimental because it leads to suboptimal control and end-organ damage, many clinicians may choose TI to minimize hypoglycemic risk. In this paper, we define TI as the failure to intensify therapy for patients with above-target HbA1c. This study sought to assess (1) the longitudinal association between TI and subsequent hypoglycemia, (2) patient and treatment factors influencing the association between TI and hypoglycemia, and (3) whether CGM use modifies the TI–hypoglycemia relationship. We hypothesized that higher TI would decrease hypoglycemia risk in the subsequent year by reducing exposure to glucose-lowering medications, and that CGM would provide a protective effect, thereby enabling safer treatment intensification with fewer hypoglycemic episodes.

2. Method

2.1. Patients, Setting, and Data Sources

We identified adults (age 18 years or older) with T2D observed from June 1, 2017, to May 31, 2023, using a previously validated method [14] based on International Classification of Diseases (ICD) 10 codes, lab results, or diabetes medications extracted from electronic health record (EHR) from two health systems in the US Midwest. The diabetes medications included orders from outpatient, urgent care, and telephone encounters: biguanides, sulfonylureas, alpha-glucosidase inhibitors, thiazolidinediones (TZD), meglitinide, glucagon-like peptide-1 agonist (GLP1), amylinomimetics, dipeptidyl peptidase-4 inhibitors (DPP4), sodium-glucose co-transporter-2 inhibitors (SGLT2), insulin, and their combinations. The final analytic cohort included 65,983 participants (see Figure 1 for cohort assembly details) that met eligibility of (1) at least one above-target HbA1c during follow-up, (2) a diabetes medication prescribed at least once during follow-up, and (3) the participant entered the cohort at least one year prior to the data cutoff date of May 31, 2023. The current-year TI was coded as “missing” in years when no HbA1c measures were above-target; subsequent-year hypoglycemia visit counts were coded as “missing” for years when participants had no encounters with the healthcare system.

Figure 1:

Cohort Assembly Details Including Data Flow of HbA1c Results, Diagnosis, and Prescription Orders

2.2. Definition of Therapeutic Inertia

The main phenomenon of interest in this analysis was TI in the current year, defined as the percentage of clinicians not adjusting drug treatment in response to above-target HbA1c and calculated as $[1-\frac{c}{H}]$ where $H$ is the number of visits (lab encounter) with an above-target HbA1c in that year, and c is the number of visits (outpatient, urgent care, and telephone encounters) in that year in which a prescription change was made [1]. In other words, TI represents the percentage of above-target HbA1c results that did not prompt a prescription adjustment within 30 days. The cutoff of above-target HbA1c is >7% for age 18–64 and >8% for age 65, since older adults may have complex or intermediate health conditions, justifying a less stringent goal. TI components were operationalized as follows:

• Qualifying event date was the date of the above-target HbA1c test.

• Current prescriptions were medications ordered in the 90-day period before the above-target HbA1c test (qualifying event).

• New prescriptions were medications ordered within 30 days of the above-target HbA1c test (qualifying event date).

• A change in prescription occurred when the current prescription differed from the new medication in terms of drug class, generic name, or total daily dose.

TI was quantified in each calendar year as the proportion of above-target HbA1c measurements for which no change in drug class, generic name, or total daily dose occurred within the subsequent 30 days (TI ranged from 0% –100%, 0 is no TI that all above-target HbA1c had an association action while 100% is complete TI that none of the elevated HbA1c were acted on). TI was coded as missing in calendar years where a participant had no above-target HbA1c values. To simplify the analyses, we focused on understanding “high TI,” that is, TI>50% in the current year. This indicates that a prescription change occurred in response to fewer than half of the above-target HbA1cs. Our definition of above-target HbA1c followed ADA guideline (>7% for ages 18–64; >8% for ages ≥65) accounting for individualized targets based on patient age and health status [15]. This reflects clinical practice to balance glycemic control with safety considerations for hypoglycemia risk and treatment-related complications in older, more frail adults. The 30-day window for prescription changes reflects real-world clinical workflows where treatment adjustments typically occur at laboratory result review via electronic prescribing, telephone, or patient portal messaging rather than exclusively at in-person visits. Although this may classify some planned delays as inertia, it captures the timeframe in which clinicians typically respond to abnormal values in contemporary practice.

2.3. Hypoglycemia Outcomes

The primary clinical adverse outcome was hypoglycemia in the subsequent calendar year (rolling 12-month periods after TI) defined as a visit by the patient to an outpatient, urgent care, or emergency department, or hospitalization for low glucose. We used ICD 10 to identify visits linked with hypoglycemia diagnosis including diabetes mellitus due to underlying conditions with or without hypoglycemia, unspecified hypoglycemia, drug-induced hypoglycemia, and poisoning by insulin and oral hypoglycemic drug accidental/ intentional [16].

2.4. Predictors of Therapeutic Inertia

Potential predictors of TI were age, gender, race (White vs. non-White), ethnicity, insurance type, comorbidity, common comorbidity (chronic kidney disease, depression, hypertension, and ischemic vascular disease), CGM use, clinic location (urban or rural), and neighborhood deprivation at the patient’s zip code. Comorbidity was measured by the Charlson Comorbidity Index (CCI) [17] and categorized as 0, 1–2, 3–4, 5–6, and >6, which corresponds to increasing risk of death [18]. The presence of common comorbidities was defined using phenotyping methods for CKD [19], depression [20], hypertension [21], and ischemic cardiac or cerebral vascular disease [22,23]. Zip codes mapped to Rural-Urban Commuting Areas Codes (RUCA) Data (v.2) were used to classify the clinic location into metropolitan, micropolitan, small town, and rural areas [24]. Neighborhood deprivation was measured using social deprivation index (SDI), an estimate of social determinant of health factors developed by Butler et al. [25] that consists of seven demographic characteristics: the percent living in poverty, single-parent households with fewer than 12 years of education, living in rental housing, living in overcrowded housing, households without a car, and unemployed adults under 65 years of age [26]. CGM use was recorded as “yes” or “no” for CGM devices (Guardian, Dexcom G4 to G6, Freestyle Libre, or Medtronic MiniMed with Guardian sensor), including devices, readers, sensors, or receivers, ordered the same year as TI.

2.5. Statistical Analysis

Summary statistics described the study's sample characteristics. Regression analyses included participant-years with non-missing data for both current-year TI and subsequent-year hypoglycemia visit counts. Logistic regression with generalized estimating equations (GEE) and an exchangeable working correlation structure were used to account for within-participant correlation across multiple years. For Aim 1 and Aim 2, we used regression to estimate the association between current-year high TI (TI >50%) on subsequent-year hypoglycemia, adjusting for patient factors (demographics, clinic rurality, and neighborhood deprivation) and treatment factors (comorbidities, insurance type, insulin therapy, and the number of diabetes drug classes in the first year of above-target HbA1c). For Aim 3 on whether CGM modifies the TI–hypoglycemia relationship, we used the same model setup for Aims 1 and 2 but added an interaction term between high TI and CGM use in the current year. Statistical analyses were conducted using R version 4.3.3 [27] plus the geepack package (version 1.3.13) for GEE model fitting with a cutoff of alpha = 0.05 for statistical significance. This study was approved by the Indiana University Institutional Review Board review #12232.

3. Results

As shown in Table 1, a total of 65,983 participants met the eligibility criteria, with a mean follow-up duration of 38 months (3.2 years). The mean age was 56 years, 24% were ≥ 65 years, half were males (51%), and most were White (75%). Among all participants, 41% had high comorbidity (CCI >5), which was more common in those ≥ 65 years old (64%). Nearly half of patients (45%) had Medicaid insurance, and 41% had Medicare coverage. Patients had an average of 1.3 HbA1c tests per year. Over the 6-year follow-up period, patients had a mean of 5 HbA1c tests (median of 3). The first HbA1c measurement averaged 8.6%, and the last measurement averaged 8.1%. Year 6 CGM data were excluded from reporting on HbA1c due to absence of year 7 hypoglycemia outcomes. More than two-thirds (68%) of the HbA1c results during the 5-year study period was above-target. CGM users had a mean above-target HbA1c percentage of 85%, compared to 67% among non-users. Adults aged 18–64 had above-target HbA1c occurrences of 74%, compared to 51% among those aged ≥ 65 years. Overall, 13% of patients had a subsequent year hypoglycemia visit (mean = 1.68 per person) (Table 2).

Table 1:

Demographic and Clinical Characteristics of the Study Cohort of Adults with T2D Observed from June 1, 2017, to May 31, 2023

	Overall	CGM Order		Age Group
		No	Yes	18-64	65+
Cohort size (n, %)	65,983	61,285 (93)	4,698 (7)	49,794 (76)	16,189 (24)
Entry time (n, %)
2017	33,971 (52)	31,515 (51)	2,456 (52)	24,520 (49)	9,451 (58)
2018	12,982 (20)	12,072 (20)	910 (19)	9,820 (20)	3,162 (20)
2019	7,410 (11)	6,962 (11)	448 (10)	5,936 (12)	1,474 (9)
2020	7,011 (11)	6,479 (11)	532 (11)	5,707 (12)	1,304 (8)
2021	4,609 (7)	4,257 (7)	352 (8)	3,811 (8)	798 (5)
Months of follow-up (mean, SD)	38 (21)	38 (22)	48 (19)	38 (21)	38 (22)
Age (mean, SD)	56 (14)	56 (14)	48 (14)	50 (11)	73 (7)
Age group (n, %)
18-44	12,867 (20)	11,189 (18)	1,678 (36)	12,867 (26)	0 (0)
45-54	15,703 (24)	14,400 (24)	1,303 (28)	15,703 (32)	0 (0)
55-64	21,224 (32)	19,974 (33)	1,250 (27)	21,224 (43)	0 (0)
65+	16,189 (24)	15,722 (26)	467 (10)	0 (0)	16,189 (100)
Male (n, %)	33,893 (51)	31,757 (52)	2,136 (46)	25,760 (52)	8,133 (50)
Race (n, %)
White	49,394 (75)	46,066 (75)	3,328 (71)	35,907 (72)	13,487 (83)
Black	12,527 (19)	11,373 (19)	1,154 (25)	10,478 (21)	2,049 (13)
Asians	1,175 (2)	1,115 (2)	60 (1)	920 (2)	255 (2)
Unknown	2,887 (4)	2,731 (4)	156 (3)	2,489 (5)	398 (2)
Hispanic (n, %)	5,041 (8)	4,751 (8)	290 (6)	4,507 (9)	534 (3)
Charlson score (n, %)
0-2	23,398 (36)	22,002 (36)	1,396 (30)	20,846 (42)	2,552 (16)
3-4	15,561 (24)	14,238 (23)	1,323 (28)	12,400 (25)	3,161 (20)
5-6	10,956 (17)	10,049 (16)	907 (19)	7,488 (15)	3,468 (21)
7 or higher	16,068 (24)	14,996 (24)	1,072 (23)	9,060 (18)	7,008 (43)
Comorbidity (n, %)1
CKD	16,335 (25)	15,256 (25)	1,079 (23)	8,983 (18)	7,352 (45)
Depression	22,718 (34)	20,667 (34)	2,051 (44)	17,242 (35)	5,476 (34)
HTN	55,966 (85)	52,128 (85)	3,838 (82)	40,822 (82)	15,144 (94)
Ischemic vascular disease	16,893 (26)	15,791 (26)	1,102 (24)	10,203 (20)	6,690 (41)
Insurance type (n, %) 2
Commercial	2,416 (4)	2,133 (4)	283 (6)	2,341 (5)	75 (0)
Medicare	27,150 (41)	25,852 (42)	1,298 (28)	12,338 (25)	14,812 (92)
Medicaid 3	29,725 (45)	27,187 (44)	2,538 (54)	28,666 (58)	1,059 (6)
Self-pay	1,814 (3)	1,725 (3)	89 (2)	1,719 (4)	95 (1)
Others	4,878 (7)	4,388 (7)	490 (10)	4,730 (10)	148 (1)
Clinic setting (n, %) 4
Primary Care	42,356 (64)	39,626 (65)	2,730 (58)	32,739 (66)	9,617 (59)
Endocrinology	8,023 (12)	6,217 (10)	1,806 (38)	10,625 (21)	4,979 (31)
Others	15,604 (24)	15,442 (25)	162 (3)	6,430 (13)	1,593 (10)
Clinic location (n, %) 4
Metropolitan	56,944 (86)	52,583 (86)	4,361 (93)	43,714 (88)	13,230 (82)
Micropolitan	5,156 (8)	4,940 (8)	216 (5)	3,583 (7)	1,573 (10)
Small town	3,594 (5)	3,474 (6)	120 (3)	2,332 (5)	1,262 (8)
Rural	58 (0)	58 (0)	0 (0)	42 (0)	16 (0)
Missing	231 (0)	230 (0)	1 (0)	123 (0)	108 (1)
Social Deprivation Index (median, IQR)	59 [33, 84]	59 [33, 83]	61 [36, 84]	59 [35, 84]	49 [32, 71]

Notes:

¹Count of comorbidity of CKD, HTN, depression, and ischemic vascular disease.

²The insurance that a patient used the most during follow-up.

³Medicaid included low-income equivalent insurance program.

⁴For clinic setting and clinic location, the most common ones observed during follow-up are listed. Others included specialty services (e.g., cardiology and gastroenterology), diabetes education, mental health, surgery, diabetes education, preadmission, and palliative care.

Table 2:

Glycemic Assessment, Control, and Glucose-lowering Medications Prescribed During the Study Follow-up Periods

		CGM Use		Age Group
	Overall	No	Yes	18-64	≥65
Cohort size (n, %)	65,983	61,285 (93)	4,698 (7)	49,794 (76)	16,189 (24)
HbA1c count per person/yr (mean, SD)	1.3 (1)	1.3 (1)	1.5 (1)	1.3 (1)	1.4 (1)
Mean HbA1c count in 6 years (median, IQR)	5 [3, 9]	5 [3, 9]	8 [5, 13]	5 [3, 9]	6 [3, 10]
Above-target HbA1c count in 5-yr, not counting 6th yr (mean, SD)	4.3 (4)	4.0 (3)	7.6 (5)	4.6 (4)	3.3 (3)
Above-target HbA1c percent in 5 yr, not counting 6th yr (mean, SD) 1	68 (30)	67 (31)	85 (22)	74 (29)	51 (30)
First HbA1c (mean, SD)	8.6 (2)	8.5 (2)	9.4 (2)	8.7 (2)	8.4 (2)
Last HbA1c (mean, SD) 2	8.1 (2)	8.0 (2)	8.7 (2)	8.1 (2)	8.0 (2)
Maximal A1c (n, %) 3
7-7.99	11,531 (18)	11,219 (18)	312 (7)	11,531 (23)	0 (0)
8-8.99	15,940 (24)	15,286 (25)	654 (14)	9,232 (18)	6,708 (41)
9-9.99	11,328 (17)	10,570 (17)	758 (16)	7,297 (15)	4,031 (25)
10-10.99	8,059 (12)	7,368 (12)	691 (15)	5,845 (12)	2,214 (14)
11-11.99	6,278 (10)	5,642 (9)	636 (14)	4,942 (10)	1,336 (8)
12-12.99	4,441 (7)	3,968 (6)	473 (10)	3,694 (7)	747 (5)
13-13.99	3,054 (5)	2,682 (4)	372 (8)	2,564 (5)	490 (3)
>=14	5,352 (8)	4,550 (7)	802 (17)	4,689 (9)	663 (4)
Prescription (mean, SD) 4	10.1(9)	9.7 (9)	14.9 (10)	10.2 (9)	9.7 (9)
Diabetes drug class (n, %)
Biguanides	37,633 (68)	35,136 (69)	2,497 (54)	30,059 (70)	7,574 (60)
Insulin	26,696 (48)	22,593 (44)	4,103 (89)	20,143 (47)	6,553 (52)
GLP1	20,223 (36)	17,868 (35)	2,355 (51)	16,770 (39)	3,453 (27)
Sulfonylureas	18,847 (34)	17,743 (35)	1,104 (24)	13,496 (32)	5,351 (42)
SGLT2	13,613 (24)	12,203 (24)	1,410 (31)	10,901 (25)	2,712 (21)
DDP4	9,635 (17)	8,980 (18)	655 (14)	7,032 (16)	2,603 (20)
Thiazolidinedione	3,632 (6)	3,388 (7)	244 (5)	2,572 (6)	1,060 (8)
Meglitinides	271 (0)	238 (0)	33 (1)	157 (0)	114 (1)
Alpha Inhibitor	126 (0)	118 (0)	8 (0)	88 (0)	38 (0)
Amylin analogs	10 (0)	8 (0)	2 (0)	9 (0)	1 (0)
Others	3,796 (7)	3,519 (7)	277 (6)	3,061 (7)	735 (6)
Number of drug class (mean, SD) 5	2.4 (1)	2.4 (1)	2.8 (1)	2.4 (1)	2.4 (1)
Subsequent-year hypoglycemia
Count (%)	8,290 (13)	7,093 (12)	1,197 (26)	5,509 (11)	2,781 (17)
Overall Mean (SD)	1.7(1.3)	1.6 (1.1)	2.2 (1.8)	1.7 (1.3)	1.6 (1.1)
Mean per year (SD)	0.4 (0.3)	0.4 (0.3)	0.4 (0.3)	0.4 (0.3)	0.3 (0.3)

Note:

Abbreviation: GLP1, glucagon-like peptide receptor agonists; SGLT2, sodium/glucose cotransporter 2 inhibitors; DDP4, dipeptidyl peptidase 4 inhibitors.

¹Calculated by the count of HbA1c above target divided by total count of HbA1c test.

²Last HbA1c at the end of the follow-up time.

³The highest HbA1c observed from 2017 to 2023.

⁴Per patient, the cumulative number of prescriptions ordered up-to-end of follow-up.

⁵Per patient, the cumulative number of drug classes ordered up-to-end of follow-up.

Despite having the greatest number of patients with CCI >5 (64%), higher prevalence of chronic kidney disease (45%), and ischemic disease (41%), individuals aged 65 years or older achieved higher glycemic control—that is, HbA1c for 49% of these patients was in target, and the last mean HbA1c was 8.0%, than the younger group. Meanwhile, the 18–64 age group had poorer glycemic control—HbA1c for 26% of this age group was in target, and the last mean HbA1c was 8.1%, which failed to meet the target of <7%. This age group also had a consistent pattern of not reaching the HbA1c target, as reflected in the distribution of maximal HbA1c (Table 2).

Biguanides were the most frequently prescribed medication (68%), followed by insulin (48%), and the most expensive and newer drug class (GLP1), ranked third (36%) (Table 2). Across all years, most individuals used 1–2 and 3 diabetes drugs classes, while many fewer took no or 4+ medications (Figure 2). Over time, being prescribed a larger number of diabetes drugs (especially 3 and 4+) became more common, suggesting a shift toward more complicated regimens over time.

Figure 2:

Trends in the Number of Diabetes Drug classes Prescribed Across Years

Only 7% of study participants used CGM at any point during the study. Among CGM users, the vast majority (98%) were prescribed insulin. Among the 61,285 participants who did not use CGM, 22,693 were prescribed insulin (44%). Hypoglycemia visits were higher among CGM users (26%) versus non-CGM users (12%). The 18–44 age group had more hypoglycemia visits than all other age groups (p < 0.001), and their proportion of three or more hypoglycemia visits was also higher (Figure 3).

Figure 3:

Patterns of Hypoglycemia Visits Stratified by Age Groups

For Aim 1, examining the longitudinal association between TI and future hypoglycemia, current-year high TI (TI >50%) was associated with increased adjusted odds of subsequent-year hypoglycemia (OR = 1.74, 95% CI 1.61–1.88, p < 0.001; Table 3). For Aim 2, examining patient and treatment factors influencing the association between TI and hypoglycemia, CGM use was associated with increased adjusted odds of subsequent-year hypoglycemia (OR = 2.26, 95% CI 1.99–2.57, p < 0.001). Age showed a strong inverse association. Older adults had substantially lower odds of hypoglycemia: ages 45–54 (OR = 0.57), ages 55–64 (OR = 0.54), and ages ≥65 (OR = 0.63; all p < 0.001) compared to younger adults age 18–44. White patients had higher odds than non-White patients (OR = 1.23, p < 0.001). Medicare insurance was associated with increased hypoglycemia risk (OR = 1.47, p < 0.001), whereas other or self-pay insurance types were not. Living in a non-metropolitan location (OR = 0.77, p < 0.001) was associated with lower odds of hypoglycemia. Higher comorbidity burden also increased the risk of hypoglycemia (Charlson Score OR = 1.12, p < 0.001). Among treatment factors, being on insulin increased the risk of hypoglycemia (OR = 2.95, p < 0.001), whereas taking more diabetes drugs lowered the risk (OR = 0.66, p < 0.001). For Aim 3, examining whether CGM modifies the TI–hypoglycemia relationship, the interaction term was not significant (95% CI: 0.47–1.1, p=0.12; Supplement Table 3), although interpretation is limited by low CGM uptake (7%) and confounding by indication.

Table 3:

Multivariable Associations Between High Therapeutic Inertia in Current Year and Hypoglycemia Visit in Subsequent Year

Variable (adjusting for entry year)	Adjusted Odds Ratio	95% CI	p-value
High TI (Ref: Low TI)	1.74	1.61, 1.88	<0.001
Age group (Ref: 18-44)
45-54	0.57	0.51, 0.62	<0.001
55-64	0.54	0.49, 0.59	<0.001
65+	0.63	0.59, 0.70	<0.001
Sex (Ref: Female)
Male	0.95	0.89, 1.00	0.072
Race (Ref: Non-White)
White	1.23	1.14, 1.33	<0.001
Neighborhood deprivation: (Ref: SDI < = 50)
SDI > 50	1.03	0.97, 1.09	0.400
Insurance (Ref: Medicaid)
Medicare	1.47	1.35, 1.59	<0.001
Others	1.13	1.02, 1.25	0.022
Self-pay	0.84	0.66, 1.06	0.150
Charlson Score	1.12	1.11, 1.12	<0.001
Clinic rurality RUCA2 (Ref: Metropolitan)
Non-Metropolitan	0.77	0.70, 0.83	<0.001
Prescribed Insulin (Ref: No)
Yes	2.95	2.71, 3.21	<0.001
Count of drug class 1	0.66	0.63, 0.70	<0.001
CGM Use (Ref: No)	2.26	1.99, 2.57	<0.001
Sulfonylureas (Ref: No)	1.04	0.91, 1.18	0.600
Hypoglycemia event in current year(Ref: No)	6.04	5.52, 6.62	<0.001

Note:

Abbreviation: CGM, continuous glucose monitoring; SDI, social deprivation index; TI, therapeutic inertia; RUCA2, Rural-Urban Commuting Areas Codes Data Version two. High TI has TI >50%, whereas low TI has TI ≤50%. Non-metropolitan included micropolitan, small towns, and rural areas.

¹Per patient, the cumulative number of drug classes ordered up to the end of follow-up. The estimated correlation parameter was 0.034 with standard error of 0.008 from GEE.

Sensitivity analyses explored reverse causality, temporal dynamics, and methodological robustness. A reverse causality model confirmed bidirectional associations: current-year hypoglycemia predicted subsequent-year high TI (adjusted OR 1.70; 95% CI 1.54–1.87; p<0.001; Supplement Table 1), indicating that prior hypoglycemia appropriately prompts clinical caution while high TI simultaneously increases hypoglycemia risk. Analysis of 83,255 person-year transitions revealed that TI is dynamic rather than fixed: 42% remained persistently high, 22% persistently low, 17% showed intensification, and 19% showed de-escalation (Supplement Table 2). These patterns reflect complex, time-varying clinical decision-making influenced by patient complexity and hypoglycemia history. Finally, using 90-day and 60-day windows instead of 30 days yielded consistent results (Supplement Tables 4-5), supporting the robustness of our primary findings.

4. Discussion

In contrast to our initial hypothesis, this large, multisite longitudinal cohort study reveals critical insights into the relationship between TI and hypoglycemia risk in diabetes management. We found that about one in eight patients (13%; Table 2) had a hypoglycemia event serious enough to require an outpatient, emergency department, or hospital visit during the 6-year study period. High TI in the present year significantly increased the adjusted odds of subsequent-year hypoglycemia by 74% (Table 3). CGM use (7% uptake) did not modify this relationship, although confounding by indication likely masked any protective effect. Specifically, clinicians may have preferentially prescribed CGM to patients already experiencing hypoglycemia or at higher risk for such events (98% of CGM users on insulin vs. 44% of non-users), making it difficult to isolate CGM's true benefit. CGM may prevent device-detected events but not severe episodes requiring medical visits. Additionally, HbA1c testing frequency in our cohort was suboptimal. Patients had a mean of 5 HbA1c tests (median of 3) over the 6-year period, averaging 1.3 tests per year—well below ADA guidelines, which recommend two tests annually for well-controlled patients (12 tests over 6 years) and four tests annually for uncontrolled patients (24 tests over 6 years) [28]. Our findings highlight several critical points that warrant discussion.

First, our finding that therapeutic inertia was associated with increased hypoglycemia risk reveals an important paradox. Physicians often avoid intensifying treatment when they perceive high hypoglycemia risk, as shown in a national survey where doctors were less likely to switch therapies for patients with major hypoglycemia risk factors [4]. However, our study found that high TI was actually associated with increased risk of subsequent-year hypoglycemia. This paradox suggests that physicians' well-intentioned caution about treatment intensification may be counterproductive. Prolonged delays may lead to more aggressive medication adjustments once treatment is finally changed, or patients remain inadequately monitored for hypoglycemia; both scenarios increase hypoglycemia risk. Patients experiencing high TI may have lower medication adherence or engagement with their diabetes care, which could prompt providers to withhold treatment intensification while simultaneously increasing hypoglycemia risk due to erratic medication-taking behavior.

Systemic barriers, including prior authorization requirements and step therapy protocols, limit access to newer, safer therapies. This was particularly problematic for Medicare beneficiaries during the study period, who faced high out-of-pocket costs for SGLT2 and GLP1 drugs [9]. Furthermore, TI may not solely reflect clinician decision-making. Patient-level factors such as missed appointments, poor medication adherence, and low engagement with diabetes self-management may contribute to both high TI and increased hypoglycemia risk. When patients do not attend follow-up visits or take medications as prescribed, clinicians lack opportunities to adjust therapy appropriately, creating apparent TI that is driven by patient behavior rather than clinical oversight. This shared responsibility between patients and providers underscores the need for interventions addressing multilevel barriers (system, clinical, and patient engagement).

Second and as expected, insulin therapy was a dominant risk factor for hypoglycemia, with insulin-treated patients experiencing nearly threefold higher odds of hypoglycemia (OR = 2.95; p<0.001), underscoring the common challenge clinicians face when trying to balance glycemic control and hypoglycemia risk. Our study extends the literature by including newer, safer medications (GLP1, SGLT2, and DPP4) and demonstrating that across all adult age groups, insulin was associated with threefold higher odds of hypoglycemia compared to non-insulin therapy. This aligns with prior studies showing elevated risk among adults aged ≥65 [29] and a sixfold increase in severe hypoglycemia requiring hospitalization [30]. A large study of approximately 1.7 million patients followed from 2006 to 2013 compared hypoglycemia rates across age groups. Among insulin users aged 45–64 compared to non-insulin users aged ≥65, severe hypoglycemia rates increased slightly in the youngest group age 18–44 (0.8 to 0.9 events per 100 person-years; p=0.025) and middle-aged group age 45–64 (0.6 to 0.9 events per 100 person-years; p<0.001) [31].

Third, important disparities surfaced across both insurance type and age groups. Medicare beneficiaries had 31% higher odds of hypoglycemia compared to Medicaid recipients, despite older adults (≥65 years) having substantially lower odds of hypoglycemia compared to younger adults (18–44 years). This discrepancy can be reconciled by recognizing that Medicare coverage extends beyond the elderly to include younger individuals with end-stage renal disease (ESRD), a population particularly vulnerable to hypoglycemia. Patients with ESRD face multiple challenges in diabetes management, including impaired renal clearance of medications, altered insulin metabolism, and the complexities of managing diabetes in the context of dialysis [32]. These challenges are magnified among insulin users. Insulin dosing becomes particularly difficult in patients on dialysis due to unpredictable insulin requirements, variable nutritional intake, and fluctuating fluid status that affects insulin sensitivity. Indeed, our cohort showed that 45% of adults aged ≥65 years had chronic kidney disease, highlighting the substantial complex decision-making in treatment intensification for diabetes. Our study extends prior research by examining insurance type as a contributing factor to hypoglycemia risk rather than focusing solely on clinical variables.

Fourth, younger adults (aged 18–44) experienced a disproportionately higher burden of hypoglycemia compared to all other age groups. They not only had more hypoglycemia-related visits when compared to all other age groups (p<0.001; Table 3) but also experienced more frequent recurrent episodes (a higher proportion having three or more visits for hypoglycemia; Figure 3). This finding aligns with prior evidence showing that the majority of hypoglycemia events requiring medical care occurred in adults aged 18–64 [33], with adults in this age range spending substantially more time below target glucose range [34]. These hypoglycemic events carry significant economic consequences, with medical costs and production losses accounting for half of all hypoglycemia-related expenditures [33]. Our findings reinforce that early to mid-adulthood represents a high-risk period for hypoglycemia that warrants tailored prevention and management strategies.

Fifth, we could not determine whether CGM mitigated hypoglycemia risk due to low uptake or confounding by indication. The very low CGM use (7%) in our cohort during 2017–2023 reflects substantial adoption barriers. Most notably, Medicare coverage was restricted to patients using three or more daily insulin injections or insulin pumps until April 2023—just one month before our study ended—when coverage expanded to all insulin-treated patients with diabetes or those with hypoglycemia [35]. This limited coverage explains the low uptake, with early adopters predominantly being patients who could afford out-of-pocket costs or had private insurance. CGM use was not associated with lower hypoglycemia visits, contrasting with type 1 diabetes trials [36,37]. However, confounding by indication—where higher-risk patients received CGM—likely masked protective effects. The low CGM uptake (7%) and substantial differences between users and non-users further complicate interpretation of this null finding. Our real-world T2D cohort differs substantially from these controlled trial settings, and our older population may have faced barriers including cognitive or visual impairments, alarm fatigue, and limited support for interpreting CGM data [38]. Further research is needed to understand factors that influence CGM effectiveness in reducing hypoglycemia in real-world T2D populations.

Limitations

This study has important limitations stemming from its observational design and EHR data. Our hypoglycemia outcome relied on healthcare encounters rather than CGM-measured glucose, likely underestimating true incidence. Additionally, 43% of person-years had missing data, though inverse-probability weighting yielded comparable results (Supplement Table 6). Unmeasured confounders—including disease severity, medication adherence, and health literacy—preclude causal inference. Our sensitivity analyses revealed bidirectional influences: hypoglycemia predicts subsequent TI just as TI predicts subsequent hypoglycemia, likely reflecting shared factors such as disease complexity and patient engagement rather than direct causation.

Distinguishing appropriate clinical restraint from problematic TI remains challenging. Our 30-day prescription window may misclassify planned delays, and undocumented hypoglycemic episodes detected outside clinical encounters are absent from EHR data, potentially misclassifying appropriate caution as inertia. Although we used age-stratified HbA1c targets and adjusted for comorbidity, we lacked data on frailty, life expectancy, and patient preferences. CGM findings (7% uptake during restrictive Medicare coverage, 2017–2023) warrant cautious interpretation given limited data on utilization patterns and substantial evolution in contemporary practice. Finally, our predominantly White Midwest population limits generalizability.

Implications

Our findings reveal a critical paradox: high TI was associated with 74% greater odds of hypoglycemia requiring a medical visit. Clinicians may delay treatment changes to avoid hypoglycemia risk, but this inaction appears to worsen the very outcome they seek to prevent. Understanding what drives TI—whether clinical caution, system barriers, or inattention to poor control—is essential for developing targeted interventions. Decision support tools that account for patient-specific factors such as kidney disease, insulin use, and hypoglycemia history could help clinicians distinguish between appropriate caution and problematic delays. However, low CGM uptake limited our ability to assess its effect on hypoglycemic events. Future research should focus on strategies to leverage CGM data for treatment intensification that improves glycemic control while minimizing hypoglycemic risk.

5. Conclusion

This large, multisite longitudinal cohort study found that high TI was common and increased the odds of subsequent hypoglycemia visits. Low CGM uptake (7%) during the study period limited assessment of its protective effect. Further research is needed to understand the drivers of TI and hypoglycemia events and to develop more effective strategies for leveraging CGM to improve diabetes outcomes.