Association between medication regimen complexity and glycemic control among patients with type 2 diabetes

Abstract

Introduction:

Type 2 diabetes mellitus (T2DM) and comorbid conditions require patients to take complex medication regimens. Greater regimen complexity has been associated with poorer T2DM management; however, the relationship between overall regimen complexity and glycemic control is unclear.

Objectives:

Our objectives were: (1) to examine associations between regimen complexity (with the Medication Regimen Complexity Index [MRCI]) and glycemic control (A1C), and (2) to compare overall MRCI with other measures of regimen complexity (overall and diabetes-specific medication count) and diabetes-specific MRCI.

Methods:

This was a secondary data analysis of cross-sectional data from a parent trial. Participants were patients with T2DM taking at least 3 chronic medications followed in safety net clinics in the Chicago area. The MRCI measures complexity based on dosing frequency, route of administration, and special instructions for prescribed medications. MRCI scores were created for overall regimens and diabetes-specific medications. Sociodemographics and outpatient visit utilization were included in models as covariates. Linear regression was used to examine the associations between variables of interest and hemoglobin A1C.

Results:

Participants (N = 432) had a mean age of 56.9 years, most were female (66.0%), and Hispanic or Latino (73.3%). Regimen complexity was high based on overall medications (mean = 6.6 medications, SD: 3.09) and MRCI (mean = 21.4, SD: 11.3). Higher diabetes-specific MRCI was associated with higher A1C in bivariate and multivariable models. In multivariable models, overall MRCI greater than 14, fewer outpatient health care visits, male gender, and absence of health insurance were independently associated with higher A1C. The variance in A1C explained by MRCI was higher compared to medication count for overall and diabetes-specific regimen complexity.

Conclusions:

More complex regimens are associated with worse A1C and measuring complexity with MRCI may have advantages. Deprescribing, increasing insurance coverage, and promoting engagement in health care may improve A1C among underserved populations with complex regimens.

Recent estimates from the U.S. Center for Disease Control indicate that about 21 million Americans are currently diagnosed with ype 2 diabetes mellitus (T2DM).¹ Glycemic control, measured by hemoglobin A1C, is one of the primary metrics used to monitor T2DM and higher A1Cs indicate worse disease management.² To lower A1C, patients are often required to engage in multiple lifestyle changes, routinely monitor blood sugar, and take medication appropriately.³

In addition to the burden of T2DM management, patients with T2DM deal with multiple chronic conditions. Nearly 86% of U.S. adults with T2DM are diagnosed with at least 2 comorbid chronic conditions, and according to systematic reviews, the average regimen size ranges from 4–10 prescription medications daily.^4,5 Medication count and polypharmacy, typically defined as 5 or more medications, are common and simple to calculate measures of regimen complexity.⁶ However, there are limitations to using simple medication count as an indicator of regimen complexity. Simple indicators fail to account for additional medication specifications which have been shown to contribute to difficulty adhering to prescribed regimens. These additional aspects of complexity include prescription instructions with higher dosing frequency (e.g., twice daily), complex routes of administration (e.g., inhalers or injections), and additional instructions (e.g., take with food).^7–11 Each of these aspects of complexity have been cited as barriers to disease self-management among patients with T2DM.⁴

Among patients with T2DM, polypharmacy is an established risk factor for poor intermediary health outcomes such as glycemic control.¹² Fewer studies have examined the associations between more comprehensive definitions of regimen complexity, as measured by validated tools like the Medication Regimen Complexity Index (MRCI), with glycemic control.^13–16 Although the MRCI is a more multi-faceted measure of regimen complexity, it is currently unclear whether it offers significant advantages over medication count alone.

Additionally, health disparities among patients with T2DM have become a major threat to public health.¹⁷ Adults belonging to racial and ethnic minority communities and those with limited English proficiency (LEP) have disproportionately worse glycemic control.^18,19 In addition to being underrepresented in medical literature in general, specifically studies of MRCI and A1C in the United States to date have been limited by being conducted at a single clinic site or among a mostly Non-Hispanic, White population.^13,16 Therefore, this study sought to fill a gap in the literature by examining regimen complexity and glycemic control in a traditionally underserved population.

Objective

The primary aim of this analysis was to examine if higher regimen complexity was associated with higher A1C in a primarily Hispanic and majority Spanish-speaking population of adults diagnosed with T2DM, with limited health literacy, and taking 3 or more medications. This study examined regimen complexity using medication count and the MRCI for all the medications in patients’ regimen and for antidiabetic medications only.

Methods

Design

This was a cross-sectional, secondary data analysis using data collected from the baseline interview of a trial examining the feasibility and effectiveness of an electronic health record-enabled intervention intended to support safe and effective prescription of drug use. Briefly, patients in 2 intervention groups were eligible to receive educational information to enhance patient knowledge of medications, a tailored medication schedule to support the consolidation of medication dosing schedules into 4 or fewer times per day, and gave clinicians access to evidence-based standardized medication label instructions in the electronic health record when prescribing medications. One of the intervention groups received an additional element consisting of 1 week of text messages to support adherence. The third group received treatment as usual. Additional details of the parent study and intervention are accessible at clinicaltrials.gov (NCT02248857).²⁰ Data were collected between December 2014 and December 2016.

Participants

Participants were recruited from 2 safety net community health centers, representing 11 primary care clinics in Chicago, Illinois. Patients completing the study were 21 years of age or older, diagnosed with T2DM, spoke English or Spanish, owned a cell phone with text message capabilities (required for intervention), took at least 3 prescription medications, and were responsible for administering their own medication. Individuals with visual, auditory, or cognitive impairments were excluded. A sample of 3092 potentially eligible patients meeting basic study eligibility criteria (i.e., age, diagnosis, number of medications) was identified based on chart review. Research staff contacted potentially eligible patients and 654 were deemed ineligible based on the application of remaining eligibility criteria (i.e., language, cell phone, administering own medications, and sensory or cognitive impairments), 502 refused, 338 passively declined, 766 were unable to be reached, and 20 were deceased before contact. After assessing eligibility and during the consent process, patients were informed that their participation in the study was dependent on their next medical visit because of the nature of the intervention, and that not all consented patients would be enrolled in the study. In total, 812 patients agreed to participate and were consented. The cooperation rate was 55.1%. Study enrollment was triggered by a recent clinic visit; therefore, only a subset of consented patients (n = 452) were enrolled in the study and randomized. After study completion, 20 participants were determined to have been taking less than 3 prescription medications at the baseline interview; therefore, making them ineligible based on the study criteria.

Procedure

Weekly chart reviews were used to identify potentially eligible patients who were then mailed letters with trial information and instructions to opt out of being called for screening. Trained bilingual research assistants called eligible patients, confirmed eligibility, and obtained verbal consent. Due to the delivery of the intervention, consented patients were not enrolled as participants until they had a clinic visit which resulted in a new prescription or change in existing prescription. The baseline interview occurred about 1 week after the clinic visit.

Approval for the trial and subsequent analyses was obtained from the Northwestern University Institutional Review Board.

Measures

Data from baseline structured interviews of the parent trial were used in this analysis. During interviews, participants were asked to list the names and verbatim instructions for all of their prescribed medications. The primary independent variable was medication regimen complexity which was evaluated in 2 ways.

1. Number of medications. The sum total of medications prescribed were recorded and collapsed into tertiles based on the data distribution.

2. MRCI. The MRCI was originally validated in a sample of patients diagnosed with chronic obstructive pulmonary disease in 2004.⁷ Since then, the measure has been translated in 6 languages and utilized among various clinical populations, including but not limited to, patients with heart failure, chronic kidney disease, and psychiatric conditions.^21–28 The MRCI consists of 3 tables of the possible dosage forms (section A), dosing frequencies (section B), and additional instructions (section C). Each item in the table corresponds to a weighted score. The calculation of complexity is based on the medication type and prescription instructions. More complex routes of administration and instructions correspond to higher weights. For example, a medication prescribed in a tablet form receives a score of 1, whereas prefilled injections correspond to a score of 3 in section A. Similarly, a medication prescribed once daily corresponds to a score of 1, whereas a medication prescribed 3 times daily corresponds to a score of 3 under section B. For section C, medications instructed to be taken before a meal would receive a score of 1. A composite complexity index is obtained by summing the score for each section and each medication. The lowest possible score on the MRCI for an individual taking at least one medication is 1.5 and there is no upper scoring limit. There are no established clinical cutoffs for the MRCI.

After completion of data collection, data analysts coded participant responses according to the MRCI instructions. In this analysis, only medications prescribed by a clinician were included in the calculation of regimen complexity. Medicinal products purchased over-the-counter were excluded; however, those that were available over-the-counter but had been prescribed by clinicians were included. Both measures of regimen complexity were calculated using antidiabetic medications only (diabetes-specific) and using all the medications in the participants’ regimen (overall). MRCI values were divided into 4 quartiles based on the data distribution. Example regimens from these quartiles are shown in Figure 1.

Figure 1.

Sample medication regimens with Medication regimen complexity index score. Abbreviations used: Medication regimen complexity index.

Baseline interviews also collected sociodemographic and health-related background information including age, race/ethnicity, education level, income, insurance type, English proficiency, health literacy, and health status. Participants were classified has having LEP if they responded to the question “How would you describe your ability to speak and understand English?” with very poor, poor, or fair. Limited health literacy was determined using the brief subjective health literacy screener.²⁹ Participants who answered somewhat, a little bit, or not at all to the question “How confident are you filling out forms by yourself” were classified as having limited health literacy. Outpatient health care utilization was examined as the frequency of medical visits in the past 6 months.

The outcome of interest was glycemic control, measured by A1C. Higher A1C levels indicate worse glycemic control and T2DM management. The closest A1C value available from 6 months prior or 3 months after the baseline interview was abstracted from the medical record.

Analysis

Descriptive statistics were calculated and reported for sociodemographic variables and covariates. Spearman correlations were used to examine the association between number of medications and MRCI. Bivariate analyses examined the association among sociodemographic characteristics, regimen complexity, and other covariates with A1C. Only variables with associations identified as statistically significant (P < 0.05) in the bivariate analyses were included in the multivariable model, with the exception of parent study intervention arm and the date difference between baseline interview and A1C value which were included as control variables. Linear regression was utilized for both the bivariate (unadjusted) and multivariable (adjusted) analyses, and variables in the multivariable model were added simultaneously. Cases with missing data were omitted. Beta coefficients (Bs) with 95% confidence intervals (CIs) are presented. Model parameters with P values and adjusted R² are reported to examine model performance of number of medications compared to MRCI. Analyses were performed using STATA version 16.1.³⁰

Results

Characteristics of the sample from this secondary data analysis are available in Table 1. The mean age of the 432 participants was 56.9 years (standard deviation [SD]: 9.47) and the majority were female (66.0%). Most participants were Hispanic or Latino (73.3%) or non-Hispanic Black (17.7%). The sample was diverse with respect to educational attainment, with about half of participants receiving less than a high school degree (fifth grade or lower: 13.7%; sixth to eighth grade: 23.6%; and ninth to 11th grade: 11.3%). Most had low income, with about half (46.5%) earning less than $15,000 per year and 41.6% earning between $15,000 and $30,000 per year. Many participants had no health insurance (34.1%) while others had Medicaid (24.8%), Medicare (17.9%), both Medicare and Medicaid (11.9%), or private insurance (11.2%). The majority of participants had LEP (58.6%) and limited health literacy (88.5%).

Table 1.

Sample characteristics (N = 432)

Variable	n (%)
Age, mean (SD)	56.9 (9.47)
Age, y
< 50	101 (23.4)
50–63	230 (53.2)
64+	101 (23.4)
Gender
Female	285 (66.0)
Male	147 (34.0)
Race and ethnicity
Hispanic or Latino	315 (73.3)
Non-Hispanic Black	76 (17.7)
Other	39 (9.1)
Education
Fifth grade or lower	59 (13.7)
Sixth to eighth grade	102 (23.6)
Ninth to 11th grade	49 (11.3)
HS graduate/GED	114 (26.4)
More than HS degree	108 (25.0)
Household income
Less than $15,000	183 (46.5)
$15,000-$30,000	164 (41.6)
More than $30,000	47 (11.9)
Insurance
Self-pay/none	143 (34.1)
Medicaid	104 (24.8)
Medicare	75 (17.9)
Medicare + Medicaid	50 (11.9)
Private/HMO	47 (11.2)
English proficiency
Limited	253 (58.6)
Proficient	179 (41.4)
Health literacy
Limited	363 (88.5)
Adequate	47 (11.5)
Outpatient visit utilization (past 6 mo)
1	88 (20.4)
2	162 (37.6)
3–4	108 (25.1)
5+	73 (16.9)

Abbreviations used: HS, high school; GED, general education development; HMO, health maintenance organization.

Note: Race/ethnicity: 2 missing, household income: 38 missing, insurance: 13 missing, health literacy: 22 missing, and outpatient visit utilization: 1 missing.

Regimen characteristics and complexity

Details of participants’ regimens are available in Table 2. Participants took a mean of 6.6 (SD: 3.1) prescription medications and 1.9 (SD: 0.96) antidiabetic medications. In terms of antidiabetic medications, 5.8% were taking none, 28.5% were taking 1, 37.5% were taking 2, and 28.2% were taking 3 or more. Less than half were prescribed insulin (40.5%), with 20.8% prescribed 1 insulin medication and 19.7% prescribed 2 insulin medications. The mean MRCI including all prescription medications was 21.4 (SD: 11.30, range: 6–83.5). The correlation between number of prescription medications and MRCI was high for overall medications (spearman rho = 0.915, P < 0.001) and diabetes-specific medications (spearman rho = 0.8816, P < 0.001).

Table 2.

Regimen characteristics

Variable		Summary value
		(N = 432)
Overall
Number of prescribed medications, mean (SD)		6.6 (3.09)
Number of prescribed medications tertile, n (%)	3–5	182 (42.1)
	6–7	128 (29.6)
	> 7	122 (28.2)
Overall MRCI, mean (SD)		21.4 (11.30)
Overall MRCI quartile, n (%)	< 14	101 (23.9)
	14–18.4	108 (25.5)
	18.5–26	97 (22.9)
	> 26	117 (27.7)
Diabetes-specific
Medication type, n (%)
Biguanides		314 (72.7)
Sulfonylureas		140 (32.4)
DPP-4 inhibitors		58 (7.3)
Thiazolidinediones		10 (2.3)
Other		12 (2.8)
Insulin (any)		175 (40.5)
Single		90 (20.8)
Multiple		85 (19.7)
Number of prescribed medications, mean (SD)		1.9 (0.96)
Number of prescribed medications tertile, n (%)	0	25 (5.8)
	1	123 (28.5)
	2	162 (37.5)
	3	105 (24.3)
	> 3	17 (3.9)
Diabetes-specific MRCI, mean (SD)		7.82 (5.04)
Diabetes-specific MRCI quartiles, n (%)	< 5	142 (32.9)
	5–7	103 (23.8)
	8–11	76 (17.6)
	> 11	111 (25.7)

Abbreviation used: MRCI, medication regimen complexity index; DPP-4, dipeptidyl peptidase-4.

Hemoglobin A1C

Overall, A1C data were available for 405 (93.8%) participants and the mean A1C of the sample was 8.15 (SD: 1.96). Over one-third (43.7%) of participants had an A1C of 8.5 or greater.

Bivariate analyses (Table 3) revealed overall MRCI was associated with A1C (< 14, mean = 7.68, SD: 1.68; 14–18.4, mean = 8.43, SD: 2.11; 18.5–26, mean = 8.43, SD: 2.16; and > 26, mean = 8.14, SD: 1.69; P = 0.02). Additional variables associated with A1C included age (P < 0.01), gender (P = 0.001), household income (P < 0.01), insurance type (P < 0.001), diabetes-specific number of medications (P < 0.001), diabetes-specific MRCI (P < 0.001), and outpatient visit utilization (P = 0.04).

Table 3.

Bivariate models examining sociodemographic, regimen, and outpatient health care utilization characteristics with A1C

Variable	A1C	P value
	Mean (SD)
Sociodemographic
Age		0.01
< 50	8.70 (2.13)
50–63	8.10 (1.95)
64+	7.73 (1.58)
Gender		0.001
Male	8.58 (2.10)
Female	7.93 (1.82)
Race and ethnicity		0.29
Hispanic or Latino	8.25 (2.00)
Non-Hispanic Black	7.87 (1.88)
Other	7.96 (1.48)
Education		0.48
Fifth grade or lower	8.19 (1.97)
Sixth to eighth grade	7.89 (1.81)
Ninth to 11th grade	7.98 (1.69)
HS graduate/GED	8.26 (2.09)
More than HS degree	8.35 (1.98)
Household income		0.01
Less than $15,000	7.93 (2.03)
$15,000-$30,000	8.13 (1.81)
More than $30,000	8.94 (1.87)
Insurance		0.001
Medicare + Medicaid	7.60 (1.76)
Medicare	7.64 (1.84)
Medicaid	7.98 (1.76)
Private/HMO	8.36 (1.73)
Self-pay/none	8.66 (2.15)
English proficiency		0.53
Proficient	8.08 (1.97)
Limited	8.20 (1.93)
Health literacy		0.13
Adequate	7.73 (1.69)
Limited	8.20 (1.98)
Medication regimen characteristics
Overall
Number of prescription medications		0.19
3–5	8.24 (2.02)
6–7	8.30 (2.03)
> 7	7.87 (1.68)
MRCI		0.02
< 14	7.68 (1.68)
14–18.4	8.43 (2.11)
18.5–26	8.43 (2.16)
> 26	8.14 (1.69)
Diabetes-specific
Number of prescription medications		0.001
0–1	6.97 (1.35)
2	8.63 (1.79)
3+	8.98 (2.06)
MRCI		0.001
< 5	6.83 (1.17)
5–7	8.53 (1.83)
8–11	8.63 (1.88)
> 11	9.17 (1.98)
Outpatient health care utilization
Outpatient visit utilization (past 6 mo)		0.04
1	8.54 (2.15)
2	8.19 (1.79)
3–4	8.09 (2.07)
5+	7.63 (1.69)

Abbreviations used: MRCI, medication regimen complexity index; HS, high school; GED, general education development; HMO, health maintenance organization.

In multivariable analyses (Table 4), the association between overall MRCI and A1C remained statistically significant for overall MRCI of 14–18.4 (B = 0.97, 95% CI: [0.40, 1.54];P < 0.01), overall MRCI of 18.5–26 (B = 1.04, 95% CI: [0.45, 1.64]: P < 0.01), and overall MRCI > 26 (B = 1.13, 95% CI: [0.53, 1.72]; P < 0.001). Outpatient visit utilization of 5 or more visits in the past 6 months was independently associated lower A1C (B = −0.92, 95% CI: [−1.53, −0.30]; P < 0.01) when compared to one visit in the same time frame. There remained a statistically significant difference in A1C based on government sponsored insurance type Medicare + Medicaid (B = −1.15, 95% CI: [−1.90, −0.39); P < 0.01), Medicare alone (B = −0.94, 95% CI: [−1.56, −0.32]; P < 0.01), or Medicaid alone (B = −0.71, 95% CI: [−1.26, −0.17]; P < 0.01), compared to participants without insurance. Male gender (B = 0.58, 95% CI: [0.16, 0.99]; P < 0.01), remained an independent predictor of higher A1C. Age was not associated with A1C in multivariable models. We tested for interactions, but these were not significant.

Table 4.

Unadjusted and adjusted multivariable models examining regimen characteristics and health-related factors with A1C

Variable	Unadjusted model		Adjusted with overall MRCI
	B (95% CI)	P value	B (95% CI)	P value
Age
< 50	-----		-----
50–63	−0.60 (−1.07, −0.14)	0.01	−0.12 (−0.65, 0.42)	0.40
64+	−0.98 (−1.52, −0.43)	0.001	−0.26 (−0.83, 0.31)	0.37
Gender
Female	-----		-----
Male	0.65 (0.25, 1.04)	0.001	0.58 (0.16, 0.99)	0.01
Household income
Less than $15,000	-----		-----
$15,000-$30,000	0.20 (−0.22, 0.61)	0.36	−0.02 (−0.46, 0.41)	0.91
More than $30,000	1.00 (0.36, 1.64)	0.01	0.53 (−0.17, 1.23)	0.14
Insurance
Self-pay/none	-----		-----
Medicare + Medicaid	−1.06 (−1.71, −0.42)	0.001	−1.15 (−1.90, −0.39)	0.01
Medicare	−1.02 (−1.57, −0.46)	0.001	−0.94 (−1.56, −0.32)	0.01
Medicaid	−0.68 (−1.79, −0.18)	0.01	−0.71 (−1.26, −0.17)	0.01
Private/HMO	−−0.30 (−0.96, 0.35)	0.37	−0.55 (−1.27, 0.17)	0.17
Overall MRCI
<14	-----		-----
14–18.4	0.74 (0.21, 1.28)	0.01	0.97 (0.40, 1.54)	0.01
18.5–26	0.74 (0.19, 1.30)	0.01	1.04 (0.45, 1.64)	0.01
>26	0.46 (−0.07, 0.99)	0.09	1.13 (0.53, 1.72)	0.001
Outpatient visit utilization (past 6 mo)
1	-----		-----
2	−0.35 (−0.86, 0.16)	0.17	−0.35 (−0.86, 0.16)	0.17
3–4	−0.45 (−1.01, 0.11)	0.11	−0.45 (−1.01, 0.11)	0.11
5+	−0.92 (−1.53, −0.30)	0.01	−0.92 (−1.53, −0.30)	0.01

Abbreviations used: HMO, health maintenance organization; B, beta coefficient; MRCI, medication regimen complexity index.

Note: Adjusted models included all variables in the table (with MRCI as the exception) and adjusted for intervention arm and number of days between baseline interview and A1C collection date.

Unadjusted and adjusted models for overall and diabetes-specific number of medications as well as overall and diabetes-specific MRCI are available in Table 5. The overall number of medications was not associated with A1C; however, number of diabetes-specific medications was associated with A1C in unadjusted (0–1: ref; 2: B = 1.66, 95% CI: [1.26, 2.06]; 3+: B = 2.02, 95% CI: [1.58, 2.45]; P < 0.001) and adjusted models (0–1: ref; 2: B = 2.02, 95% CI: [1.58, 2.45]; 3+: B = 2.09, 95% CI: [1.63, 2.56]; P < 0.001). Diabetes-specific MRCI was also associated with higher A1C in unadjusted models (< 5: ref; 5–7: B = 1.70, 95% CI: [1.25, 2.14], P < 0.001; 8–11: B = 1.80 95% CI: [1.31, 2.29], P < 0.001; and > 11: B = 2.33, 95% CI: [1.90, 2.77], P < 0.001). In adjusted models, diabetes-specific MRCI remained independently associated with higher A1C (< 5: ref; 5–7: B = 1.48, 95% CI: [1.01, 1.95], P < 0.001; 8–11: B = 1.77, 95% CI: [1.23, 2.30], P < 0.001; and > 11: B = 2.22, 95% CI: [1.74, 2.69], P < 0.001). Diabetes-specific MRCI accounted for the greatest variance in A1C (unadjusted model R² = 0.2426) followed by diabetes-specific number of medications (unadjusted model R² = 0.1763) and overall MRCI (unadjusted R² = 0.0263).

Table 5.

Model parameters for measures of regimen complexity and A1C

	Unadjusted			Adjusted
Measure of regimen complexity	B (95% CI)	P value	R2	B (95% CI)	P value	R2
Overall
Number of medications	-----		0.0082
3–5	0.06 (−0.39, 0.51)	0.80
6–7	−0.37 (−0.83, 0.10)	0.12
> 8
MRCI
< 14	-----		0.0263	-----		0.1638
14–18.4	0.74 (0.21, 1.28)	0.01		0.97 (0.40, 1.54)	0.01
18.5–26	0.74 (0.19, 1.30)	0.01		1.04 (0.45, 1.64)	0.01
> 26	0.46 (−0.07, 0.99)	0.09		1.13 (0.53, 1.72)	0.001
Diabetes-specific
Number of medications
0–1	-----		0.1763	-----		0.3038
2	1.66 (1.26, 2.06)	0.001		1.64 (1.22, 2.06)	0.001
3+	2.02 (1.58, 2.45)	0.001		2.09 (1.63, 2.56)	0.001
MRCI
<5	-----		0.2426	-----		0.3145
5–7	1.70 (1.25, 2.14)	0.001		1.48 (1.01, 1.95)	0.001
8–11	1.80 (1.31, 2.29)	0.001		1.77 (1.23, 2.30)	0.001
>11	2.33 (1.90, 2.77)	0.001		2.22 (1.74, 2.69)	0.001

Abbreviations used: MRCI, medication regimen complexity index.

Note: Adjusted models adjusted for age, gender, household income, outpatient visit utilization, intervention arm and number of days between baseline interview and A1C collection date.

Discussion

This secondary data analysis sought to understand the association between regimen complexity using patients’ full regimen and A1C among adults with T2DM taking 3 or more medications. Further, we examined this question among a unique sample in which most study participants had multiple social determinants of health including Hispanic or Latino race and ethnicity, LEP, limited health literacy, and low income. The results of this observational study indicated male sex, overall MRCI above 14, a single outpatient health care visit in 6 months, and lack of insurance were independently associated with higher A1Cs. Diabetes-specific number of medications and MRCI were also highly associated with A1C.

To our knowledge, this is the first study to link higher overall MRCI to higher A1C in a U.S. population. These findings are consistent with Ab Rahman et al.’s recent study among patients with T2DM followed in public primary care clinics in Malaysia which found MRCI was associated with reduced odds of achieving A1C ≤7.0% (adjusted odds ratio: 0.89) and ≤7.5% (adjusted odds ratio: 0.90).¹⁵ Compared to the Malaysian study sample, our sample was slightly more complex in terms of medication count (4.8 vs. 6.6 medications) and overall MRCI (15.1 vs. 21.4) although similar by diabetes-specific MRCI (7.9 vs. 7.8). Ab Rahman et al. also found that the R² of multivariable models with overall MRCI accounted for greater variance in A1C compared to overall medication count. This is consistent with our findings that overall MRCI had stronger associations with A1C than overall medication count, which in this analysis was nonsignificant. Findings from these 2 studies suggest that overall MRCI may have advantages to overall medication count alone when evaluating all the medications in the patient’s regimen and common markers of disease progression. However, given the high correlation between overall number of medications and overall MRCI, it is unclear if the benefits of measuring MRCI would result in clinically meaningful improvement in patient outcomes in practice.

In this observational study among a single cohort, overall MRCI of 14 or more was associated with a cross-sectional measure of A1C. The original validation of the MRCI did not include clinically meaningful cutoffs defining complexity; therefore, studies analyzing the MRCI categorically have created their own definitions of low, moderate, or high complexity using sample distributions. Thus, there is considerable variability due to heterogeneity between studies. However, added clinical support for these cutoffs is a study finding the mean MRCI, which was 18.0 (SD: 8.5) for solid organ transplant patients who are generally considered to have high regimen complexity.³¹ The degree to which these definitions of complexity or clinical cutoffs would be applicable in other regions of the United States and health systems is unknown and could be a direction of future research. Research studies using big-data approaches designed to examine the relationships between complexity and clinical outcomes may be warranted.

Prior research has also found that higher diabetes-specific MRCI has been associated with higher A1C.^13–15 Whereas a threshold effect was observed between overall MRCI and A1C with scores over 14 having similar magnitude of associations with A1C, a trend effect was found between diabetes-specific MRCI and A1C. This analysis did not examine or rate appropriateness of prescribing; therefore, it is possible that this increasing complexity is a result of guideline consistent treatment of disease progression. For instance, in treatment of T2DM, insulin should be “considered” in patients with A1C ≥ 7.5 and is considered “essential” for patients with A1C > 10%.³² As injectable medications have a minimum MRCI of 3.5 and tablet medications have a minimum of 1.5, with the addition of insulin, MRCI increases significantly. Therefore, the strong positive associations between diabetes-specific MRCI and A1C are expected. With increasing disease severity, clinicians must weigh the benefits and harms of medications with more complex directions for use and side effect profiles.³³ In many situations, regimen complexity may be unavoidable; however, it is also well known that polypharmacy is associated with adverse drug events and poorer adherence,³⁴ which could be mitigated by deprescribing when clinically appropriate.³⁵ Reducing complexity further by simplifying dosage frequency or prescribing combination medications should also be considered to support patients.^36–39 No current studies to our knowledge have examined if reducing MRCI leads to improvement in A1C, which could be an opportunity for future research.

This study was comprised of Hispanic or Latino and Black adults, who have increased risk of developing T2DM.⁴⁰ Health disparities among traditionally underserved populations have been attributed to a complex interaction of biological, clinical, social, and health system factors.⁴¹ No U.S. studies examining MRCI to date have examined racial or ethnic disparities in regimen complexity among adults with T2DM; therefore, it is difficult to contextualize our findings with related literature. Beyond how medications are prescribed, there are many barriers which influence the ability to take medications as prescribed and follow other guideline consistent recommendations which are disproportionately experienced by racial and ethnic minorities.⁴² For instance, prior research has shown that any insurance is associated with improved glycemic control in traditionally underserved populations,^43–45 which is consistent with our finding that any form of government sponsored insurance was associated with lower A1C when compared to no insurance. Studies suggest that this may be attributed to higher rates of blood glucose testing among the insured⁴⁶ or differing clinical management decisions made by providers based on patient coverage.⁴⁷ Additionally, this study found A1C was similar among patients with private insurance and those without insurance. Although policy changes, such as the Affordable Care Act, have increased access to insurance, the poorest individuals with private insurance may still experience significant barriers.⁴⁸ Relatedly, nearly monthly clinic visits were associated with lower A1C, which could be related to successful outreach of clinics in treating at-risk patients, more opportunities for clinical optimization, or patient factors such as higher health activation or adherence.^49,50 These findings support the health benefits of insurance coverage and engagement in care among traditionally underserved populations.

This study had some notable limitations. This was a secondary data analysis and as with any secondary data analysis, issues of the parent trial can contribute to bias in derivative studies. We attempted to reduce this bias by using data from the baseline interview only and adding intervention arm of the parent trial as a covariate in multivariable models. Also, the generalizability of our findings is limited by the parent study sample. First, this population was comprised of adults with multiple social determinants of health and is not representative of the general population. However, these populations are often underrepresented in research, making examining regimen complexity in our sample an important contribution to the literature. Second, all the participants in our study were prescribed a minimum of 3 medications, therefore, they may not be generalized to adults with simpler medication regimens. In addition, this study was cross-sectional and the main dependent variable was a single measure of A1C, so we cannot determine causality between MRCI and A1C or examine the role of potential confounding variables which may explain the association. Additionally, hypoglycemia has been increasingly targeted as a major intermediary outcome of interest among patients with T2DM, but data on hypoglycemic events were not available.

Conclusions

Higher overall MRCI, lack of health insurance coverage, less frequent outpatient health care utilization, and male gender were independently associated with higher A1C among a sample of primarily Spanish-speaking, Hispanic adults with T2DM. Although the correlations between number of prescribed medications and MRCI were high, MRCI was more strongly associated with A1C for in both overall and diabetes-specific analyses. Efforts to simplify medications, when clinically appropriate, may be warranted.

Key Points.

Background:

• Regimen complexity is associated with utilization of emergency and hospital services.

• Less is known about regimen complexity and diabetes management quality indicators (e.g., hemoglobin A1C).

Findings:

• Greater complexity of patients’ overall and diabetesspecific prescription regimen was linked to higher A1C among adults with T2DM and from traditionally underserved backgrounds.

• In this study, the medication regimen complexity index explained more of the variance in A1C than simple medication count.