A retrospective study of persistence, adherence, and health economic outcomes of fixed-dose combination vs loose-dose combination of oral anti-diabetes drugs

Abstract

Objective:

To compare outcomes between patients with type 2 diabetes mellitus (T2DM) using fixed-dose combination (FDC) and loose-dose combination (LDC) products.

Methods:

This retrospective cohort study used MarketScan® Commercial and Medicare Supplemental data from January 1, 2009–December 31, 2013. The identified population included patients with T2DM and ≥1 additional oral anti-diabetic prescription (of the same regimen [FDC/LDC] as the index prescription) within 12 months following the fill date. Persistence (no ≥30-day gap) and adherence (medication possession ratio [MPR] ≥0.8) were assessed as primary end-points; secondary end-points included hypoglycemia, healthcare resource utilization, and costs.

Results:

Of 23,361 patients identified, 12,590 (53.9%) were on FDC therapy and 10,771 (46.1%) were on LDC therapy. FDC patients had a significantly lower rate of non-persistence (67.9% vs 73.4%, p < 0.0001) and a significantly higher rate of adherence to therapy (57.0% vs 50.7%, p < 0.0001) when compared to LDC patients. Average time to non-persistence was significantly longer among FDC vs LDC patients (207.1 vs 186.3 days, p < 0.0001). After adjusting for baseline characteristics, the odds of non-persistence were 21% lower with FDC vs LDC therapy (OR = 0.79, 95% CI = 0.74–0.85, p < 0.0001), with a 28% higher odds of adherence (OR = 1.28, 95% CI = 1.20–1.36, p < 0.0001). Differences in most secondary outcomes significantly favored FDC therapy, including total predicted monthly all-cause costs ($1008 vs $1053; p = 0.006) and T2DM-related costs ($142 vs $155; p < 0.001).

Limitations:

Cohort classification was based on prescription claims data. The lack of clinical data limits assessment of potential influencers of FDC vs LDC decisions, residual confounding was possible, and diabetes-related medical costs only captured claims with a primary diagnosis for diabetes. The results may not be generalizable to populations such as Medicaid.

Conclusion:

Management of T2DM using FDC therapies provides a compliance benefit relative to LDC therapies that may translate to reductions in healthcare utilization and costs.

Keywords::

• Diabetes
• Persistence
• Adherence
• Fixed dose combination
• Costs

Introduction

Treatment of type 2 diabetes mellitus (T2DM) consists of significant lifestyle adjustments and drug therapy, including oral anti-diabetic agents (OADs) and in some cases insulin therapy. The goal of anti-diabetic therapy is to maintain glycemic control, with the goal of lowering the rates of microvascular and macrovascular complications, including neuropathy, nephropathy, retinopathy, coronary artery disease, peripheral artery disease, and stroke1. A limiting factor in achieving control is the simultaneous avoidance of hypoglycemia. Several groups of drugs (oral formulations and injectables) are effective in maintaining glycemic control in T2DM. While the American Association of Clinical Endocrinologists (AACE)/American College of Endocrinologists (ACE) diabetes management algorithm articulates that the A1c target must be individualized based on numerous factors, AACE and ACE recommend that dual therapy begins for treatment-naïve patients with an A1c in the range of 7.6–9.0% (because no single medication is likely to attain a goal of <6.5%)2.

Previous research has provided significant evidence of the advantages of fixed-dose combination (FDC) over loose-dose combination (LDC) treatment in the management of T2DM3–12. Recently, there has been discussion of positioning FDCs earlier in the treatment pathway, which has advantages from the perspective of both physicians and patients13^,14. More specifically, the advantages identified and discussed in support of FDCs for treatment of patients with T2DM are greater efficacy compared with higher dose monotherapy, reduced risk of adverse reactions relative to higher dose monotherapy, lower overall costs, and improved compliance. A recent claims-based retrospective study conducted by Williams et al.5 concluded that treatment initiation with FDC was associated with lower healthcare resources utilization and costs and better likelihood of A1c goal attainment compared to treatment initiation with LDC. The research, however, has been limited by the use of intent-to-treat approaches that do not account for switching between treatment regimens (FDC to LDC and vice versa) or discontinuation. Further, persistence and adherence were not part of their study objectives.

Given the limitations of the existing literature surrounding FDC vs LDC therapy for T2DM, we designed a study that sought to comprehensively measure and compare persistence, adherence, and economic outcomes of FDC vs LDC users, all from one single data source, using a large national US claims database with a robust sample size.

Patients and methods

Data from January 1, 2009–December 31, 2013 (latest available data) were used for this study, representing the study period. A retrospective, observational cohort study design was used, as shown in Figure 1.

Figure 1. Study design. FDC, fixed-dose combination; LDC, loose-dose combination. *Index date = date of first oral anti-diabetic prescription during enrollment period.

Adults with a prescription for at least one FDC or two LDC medications (filled on the same date) for T2DM during the enrollment period from January 1, 2010–December 31, 2012 were identified. The index date was defined as the fill date of the first OAD prescription during the enrollment period. A 12-month period before the index date (pre-index) was used to obtain a baseline assessment of the study sample. All outcomes were assessed over a maximum of 1 year following and including the index date.

This study was conducted in accordance with International Society for Pharmacoepidemiology (ISPE) Guidelines for Good Pharmacoepidemiology Practices (GPP) and applicable regulatory requirements. All patient records were de-identified and fully compliant with US patient confidentiality requirements, including the Health Insurance Portability and Accountability Act (HIPAA) of 1996.

Study population

Adults (age ≥18 years on the index date) diagnosed with T2DM, naïve to anti-diabetic treatment during the pre-index period and having at least two OAD prescriptions of the index regimen during follow-up comprised the study population. T2DM was identified by the presence of a non-diagnostic medical claim with International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9 CM] codes of 250.x0 (type 2 diabetes, not stated as uncontrolled) or 250.x2 (type 2 diabetes, uncontrolled), within 12 months following the index date. Patients were classified into two cohorts depending on the type of index OAD medication regimen: (1) FDC cohort of patients with at least two prescriptions filled for any FDC therapy beginning on the index date or (2) LDC cohort of patients with at least two prescriptions filled for two different non-FDC OAD classes on the index date with at least two distinct 15-day overlaps in any two non-FDC prescription fills during follow-up. Additional inclusion criteria included continuous enrollment in medical and pharmacy benefits throughout the pre-index and follow-up period.

Exclusion criteria included a diagnosis of pregnancy, gestational diabetes, secondary diabetes, or type 1 diabetes anytime during the pre-index and follow-up period, filled prescriptions for OADs during the pre-index period, filled prescriptions for non-oral antidiabetic medications anytime during the pre-index and follow-up period, and fills for triple therapy (either FDC + third agent or 3 agents in loose dose form) on the index date or during follow-up (two distinct 15 day overlaps in a FDC and LDC medication or three LDC medications during follow-up). Patients who received anti-diabetic monotherapy or who had ≥1 capitated medical claim anytime during the pre-index and follow-up period were also ineligible.

Patients had to have ≥1 additional OAD prescription fill (of the same regimen [FDC/LDC] as the index prescription) within 12 months following the index date. Switching to different agents in the same regimen (FDC or LDC) was not considered discontinuation of therapy for the study purposes. For example, if patients switched from one FDC product to another FDC product they were to be evaluated as being on FDC therapy. The same applied to the LDC cohort. Switching from one class of drugs to another (e.g., sulfonylureas to thiazolidinediones) within the same regimen was to be reported but not considered as discontinuation of the LDC regimen.

Patients were to be followed up to the last day of available index regimen medication (FDC or LDC) or 365 days whichever comes first. The person-years contributed by each patient in the population were summed to obtain total person-years during follow-up. All outcomes were reported in terms of person-years. The person-year approach was considered suitable for the study objectives as it accounts for switching or discontinuation of index regimen type while comparing outcomes. Therefore, a hypoglycemic event during the 1-year follow-up period, for example, would not be attributed to the index regimen type if the patient was no longer on the index regimen.

Data source

The Truven Health MarketScan® Commercial Claims and Encounters Database (Commercial) and the Truven Health MarketScan® Medicare Supplemental and Coordination of Benefits Database (Medicare Supplemental) were used for this study. The MarketScan® Commercial Database consists of employer- and health plan-sourced data containing medical and drug claims for over 40 million individuals annually. The MarketScan® Medicare Supplemental Database contains the inpatient and outpatient medical and prescription claims of Medicare-eligible persons with supplemental insurance plans offered by their former employers. There are ∼4.3 million enrollees annually included in the database. Both databases provide detailed cost, use, and health outcomes (e.g., adverse events, inpatient death) data for healthcare services performed in both inpatient and outpatient settings. Medical claims are linked to outpatient prescription drug claims and person-level enrollment information using blinded patient identifiers.

Outcomes

The primary outcomes were persistence and adherence. Persistence was defined as a binary variable, indicating any gap of ≥30 days in the index therapy regimen during the follow-up period. Medication possession ratio (MPR) was used to measure adherence as both a continuous and binary measure. The MPR numerator was defined as the sum of the days’ supply for all index regimen medication fills during the follow-up period, with a denominator of the number of days between index date and the date of the last available day of the index regimen medication dispensed during follow-up. If the days’ supply extended beyond the end of follow-up then the supply was truncated appropriately.

For dual therapy users (LDC), the dual therapy MPR12 (DTMPR) was calculated as follows:

Patients were considered to be adherent if the MPR or DTMPR was ≥0.8 (non-adherent if the MPR or DTMPR was <0.8). In addition to the binary indicator, the continuous measure of MPR or DTMPR was also reported; if MPR exceeded 1.0, it was truncated to 1.0 for the analysis.

Secondary end-points included hypoglycemia, healthcare resource utilization, and costs. For hypoglycemia, the proportion of patients with hypoglycemic events and number of hypoglycemic events per person-year were reported for each patient cohort. A hypoglycemic event was defined as a diagnosis of hypoglycemia (ICD-9 CM: 251.0x, 251.1x, 251.2x) on an outpatient or emergency room (ER) claim, a principal diagnosis on a hospital claim, or a glucagon injection in an outpatient setting15. Healthcare resource utilization included all-cause and diabetes-related (if the primary ICD-9 CM on the claim was 250.x0 or 250.x2) resource utilization, capturing hospitalizations and emergency room, physician office, laboratory, radiology, and other outpatient visits. All-cause and diabetes-related costs were measured while the patient was on index regimen using medical and prescription claims. Costs reflected all payments made to providers of care from both the plan (plan and co-ordination of benefits) and the patient (co-payment, co-insurance, deductible). All costs were adjusted to 2013 US dollars using the Bureau of Labor Statistics’ medical care component of the Consumer Price Index (CPI)16. Monthly medical, prescription (all-cause and diabetes-related prescriptions [all OADs]), and total costs were reported.

Statistical analysis

Sample size calculations were done using the G*Power program, and initial estimates of proportion persistent to FDC and LDC therapy (90-day persistence rates), respectively, were obtained from a similar study of patients with T2DM17. Additionally, in previously published studies using a similar population, the ratio of FDC to LDC patients was 2:1⁵. Assuming proportions of persistent patients of 67% for FDC vs 50% for LDC, using alpha of 0.05 and 95% statistical power, the target sample size was 432 patients for a 1:1 FDC:LDC ratio (n = 216 per group) or 483 patients for a 2:1 ratio (n = 322 FDC; n = 161 LDC). Preliminary estimates from prior studies using claims databases supported a total sample size of at least 12,641 T2DM-diagnosed treatment naïve patients initiating FDC or LDC therapy5. Such a sample size suggests that up to a 75% (432/12,641) attrition rate would still permit a statistically powered study. Therefore, we anticipated a sufficient sample size to allow assessment of differences in proportions persistent to therapy between the two study cohorts.

Descriptive statistics (means and proportions) were used to characterize the study sample overall and for each study cohort. Chi-square testing were used to evaluate differences in proportions between categorical variables across study cohorts (FDC vs LDC), whereas ANOVA/t-tests or Kruskall-Wallis testing (depending on normality) were used for differences in means of continuous variables. All multivariable analyses controlled for clinically meaningful demographic and clinical variables (e.g., age, gender, plan types, region, Charlson comorbidity index [CCI], other co-morbid conditions, pre-index resource utilization and costs). All statistical tests were based on a 2-sided hypothesis of no difference between study cohorts at a significance level of 0.05. Data management was conducted using Statistical Analysis System (SAS) 9.2 and all analyses performed using SAS 9.2 and STATA/MP 13.

For persistence and adherence, preliminary analyses tested for unadjusted differences of primary end-points between study cohorts using Chi-square and t-tests or Kruskal-Wallis tests as appropriate. Logistic regression models assessed differences in the proportion of patients persistent (% patients with no gap ≥30 days) or adherent (% patients with MPR/DTMPR ≥0.8) between the two study cohorts while controlling for baseline demographic and clinical characteristics. Time to non-persistence was analyzed according to Kaplan-Meier methods and Cox proportional hazards models (controlling for baseline characteristics). Logistic regression models were also used to obtain the adjusted differences in odds of a hypoglycemic event occurrence, hospitalization, and ER visit, respectively, between the two cohorts. Zero-inflated negative binomial regression models (based on data dispersion) were used in the assessment of the difference in the rate of health resource utilization. Differences in total, medical, and prescription costs between cohorts were assessed using generalized linear models (GLMs) with the appropriate distribution and link functions (e.g., log link and gamma error distribution) while controlling for baseline characteristics. Since the GLM estimation directly handles the skewness of cost data, there is no need for back transformation commonly used in ordinary least square models with the log transformed dependent variable. When used for modelling costs the Gamma distribution combined with the log link is the most common choice.

Results

A total of 558,123 patients were identified in the database as fitting the definition for the target population (i.e., had ≥2 prescriptions filled for any FDC therapy over 12 months beginning on the index date or ≥2 prescriptions filled for two different non-FDC OAD classes on the index date with ≥2 distinct 15-day overlaps in any two non-FDC prescription fills during follow-up). Of these, 23,361 patients (4.2% of those initially identified) were eligible for the final study population (n = 12,590 and n = 10,771 for the FDC and LDC cohorts, respectively). The most common reasons for exclusion were: non-continuous enrollment in medical and pharmacy benefits throughout the pre-index and follow-up periods (n = 297,754) and had filled prescriptions for OADs during the pre-index period (n = 294,570).

Baseline characteristics for the study population are shown in Table 1. Statistical differences between the FDC and LDC cohorts were observed in all parameters except gender, CCI, and co-morbid depression. Overall, the FDC cohort was younger, had more endocrinologist visits, and had lower co-morbidity rates with a few exceptions (namely dyslipidemia and hypertension). Regarding the most common index drug combinations, they were dipeptidyl peptidase-4 (DPP4) inhibitors plus biguanides (50.9%) and sulfonylureas plus biguanides (31.6%) with FDC and sulfonylureas plus biguanides (75.9%) and DPP4 inhibitors plus biguanides (8.3%) with LDC.

Table 1. Baseline characteristics.

Characteristic	All		FDC		LDC		p*
	(n = 23,361		(n = 12,590)		(n = 10,771)
Age in years, mean (SD)	54.9	(11.2)	54.5	(10.9)	55.3	(11.6)	<0.0001
Gender, n (%)
Male	14,581	62.4%	7790	61.9%	6791	63.0%	0.0647
Geographic region, n (%)
Northeast	3644	15.6%	2041	16.2%	1603	14.9%	<0.0001
North Central	5774	24.7%	2739	21.8%	3035	28.2%
South	10,512	45.0%	6251	49.7%	4261	39.6%
West	2350	10.1%	958	7.6%	1392	12.9%
Unknown	1081	4.6%	601	4.8%	480	4.5%
Plan type, n (%)
FFS	2225	9.5%	1105	8.8%	1120	10.4%	<0.0001
EPO/PPO	14,581	62.4%	7939	63.1%	6642	61.7%
HMO	1569	6.7%	860	6.8%	709	6.6%
POS	1896	8.1%	1092	8.7%	804	7.5%
CDHP/HDHP	1318	5.6%	697	5.5%	621	5.8%
Unknown	1772	7.6%	897	7.1%	875	8.1%
Endocrinologist visit, n (%)	804	3.4%	467	3.7%	337	3.1%	0.0153
CCI, mean (SD)	1.1	(1.2)	1.1	(1.1)	1.1	(1.3)	0.6041
Specific comorbidities, n (%)
Cerebrovascular disease	914	3.9%	411	3.3%	503	4.7%	<0.0001
Hemorrhagic stroke	33	0.1%	11	0.1%	22	0.2%	0.0177
CKD	267	1.1%	108	0.9%	159	1.5%	<0.0001
Coronary heart disease	2551	10.9%	1315	10.4%	1236	11.5%	0.0118
MI	407	1.7%	153	1.2%	254	2.4%	<0.0001
Depression	931	4.0%	484	3.8%	447	4.2%	0.2338
Dyslipidemia	8656	37.1%	4985	39.6%	3671	34.1%	<0.0001
Hypertension	11,244	48.1%	6279	49.9%	4965	46.1%	<0.0001
Obesity	1724	7.4%	886	7.0%	838	7.8%	0.0304
No. of unique prescription classes, mean (SD)	2.6	(2.9)	2.9	(2.9)	2.3	(2.8)	<0.0001
Index medication classes†, n (%)
Alpha-glucosidase Inhibitors	16	0.1%	–	–	16	0.1%	–
Meglitinides	118	0.5%	12	0.1%	106	1.0%	<0.0001
Biguanides	22,521	96.4%	12,480	99.1%	10,041	93.2%	<0.0001
DPP4 Inhibitors	7684	32.9%	6412	50.9%	1272	11.8%	<0.0001
Sulfonylureas	12,831	54.9%	4086	32.5%	8745	81.2%	<0.0001
Thiazolidinediones	3403	14.6%	2192	17.4%	1211	11.2%	<0.0001
Bile acid sequestrants	152	0.7%	–	–	152	1.4%	–
Index drug combinations, n (%)
DPP4 Inhibitors + Biguanides	7302	31.3%	6412	50.9%	890	8.3%	<0.0001
DPP4 Inhibitors + Thiazolidinediones	106	0.5%	0	0.0%	106	1.0%	<0.0001
Meglitinides + Biguanides	82	0.4%	12	0.1%	70	0.6%	<0.0001
Sulfonylureas + Biguanides	12,147	52.0%	3976	31.6%	8171	75.9%	<0.0001
Sulfonylureas + Thiazolidinediones	417	1.8%	111	0.9%	306	2.8%	<0.0001
Thiazolidinediones + Biguanides	2862	12.3%	2082	16.5%	780	7.2%	<0.0001
Bile acid sequestrant combinations	152	0.7%	N/A		152	1.4%	–
Alpha-glucosidase inhibitor combinations	16	0.1%	N/A		16	0.1%	–
Other combinations	276	1.2%	N/A		276	2.6%	–

CCI, Charlson comorbidity index; CDHP, consumer-driven health plan; CKD, chronic kidney disease; DPP4, dipeptidyl peptidase-4; EPO, exclusive provider organization; ER, emergency room; FDC, fixed-dose combination; FFS, fee-for-service; HDHP, high-deductible health plan; HMO, health maintenance organization; LDC, loose-dose combination; MI, myocardial infarction; POS, point of service; PPO, preferred provider organization; SD, standard deviation; T2DM, type 2 diabetes mellitus.

*p-values were obtained using t-tests for continuous variables and Chi square tests for categorical variables. p-values for differences in costs were obtained from the non-parametric Wilcoxon-Mann-Whitney test.

†Not mutually exclusive.

Unadjusted outcomes

Data derived from the unadjusted analyses are presented in Table 2. The FDC cohort was found to have a significantly lower rate of non-persistence (67.9% vs 73.4% for LDC; p < 0.0001) and greater average time to non-persistence (207.1 vs 186.3 days; p < 0.0001), along with a higher rate of adherence (57.0% vs 50.7%; p < 0.0001) and lower rate of hypoglycemia (0.57% vs 0.99%; p = 0.0002). In terms of resource utilization, statistical differences were seen in each outcome except for the number of radiology visits, with all-cause and T2DM-related parameters consistently favoring the FDC cohort. All-cause monthly costs were likewise lower in the FDC cohort, except for prescription costs ($290 vs $225 with LDC; p < 0.0001), which did not lead to an increase in total costs vs the LDC cohort ($935 vs $977; p < 0.0001); conversely, a significant increase in T2DM-related prescription monthly costs for FDC vs LDC therapy ($119 vs $59; p < 0.0001) contributed to significantly higher T2DM-related total monthly costs ($168 vs $120; p < 0.0001).

Table 2. Unadjusted outcomes.

Characteristic	All		FDC		LDC		p*
	(n = 23,361)		(n = 12,590)		(n = 10,771)
Non-persistent (≥30-day gap), n (%)	16,460	70.5%	8551	67.9%	7909	73.4%	<0.0001
Time to non-persistence (days), mean (SD)	197.5	(133.8)	207.1	(136.3)	186.3	(130.0)	<0.0001
Adherent (MPR/DTMPR ≥0.8), n (%)	12,637	54.1%	7176	57.0%	5461	50.7%	<0.0001
Adherence (MPR/DTMPR), mean (SD)	0.77	(0.2)	0.78	(0.2)	0.76	(0.2)	<0.0001
Hypoglycemia, n (%)	179	0.8%	72	0.57%	107	0.99%	0.0002
No. of hypoglycemic events, rate per PY	0.012		0.008		0.016		<0.0001
Resource utilization
All-cause resource utilization
Hospitalization, n (%)	2021	8.7%	929	7.4%	1092	10.1%	<0.0001
Length of stay, days per PY	0.56		0.42		0.71		<0.0001
ER visits, n (%)	4086	17.5%	1940	15.4%	2146	19.9%	<0.0001
No. of ER visits, rate per PY	0.31		0.26		0.37		<0.0001
No. of physician office visits, rate per PY	6.85		6.61		7.11		<0.0001
No. of laboratory visits, rate per PY	1.63		1.56		1.71		<0.0001
No. of radiology visits, rate per PY	0.88		0.83		0.93		<0.0001
No. of other outpatient visits, rate per PY	6.09		5.61		6.63		<0.0001
T2DM-related resource utilization
Hospitalization, n (%)	48	0.2%	18	0.1%	30	0.3%	0.0226
Length of stay, days per PY	0.02		0.01		0.03		<0.0001
ER visits, n (%)	437	1.9%	167	1.3%	270	2.5%	<0.0001
No. of ER visits, rate per PY	0.03		0.02		0.03		<0.0001
No. of physician office visits, rate per PY	2.76		2.62		2.91		<0.0001
No. of laboratory visits, rate per PY	0.82		0.81		0.84		0.0051
No. of radiology visits, rate per PY	0.01		0.01		0.01		0.1564
No. of other outpatient visits, rate per PY	1.21		1.11		1.32		<0.0001
Costs
(monthly) All-cause costs (2013 USD), mean (SD)
Total	$954	(2267)	$935	(2381)	$977	(2127)	<0.0001
Medical	$694	(2187)	$644	(2313)	$752	(2029)	<0.0001
Hospitalization	$234	(1675)	$219	(1,880)	$251	(1,398)	<0.0001
ER	$42	(238)	$38	(263)	$47	(204)	<0.0001
Physician office	$128	(457)	$123	(479)	$134	(431)	<0.0001
Laboratory	$17	(55)	$17	(57)	$17	(54)	0.0004
Radiology	$35	(196)	$33	(182)	$37	(211)	0.0040
Other outpatient	$238	(753)	$214	(722)	$266	(787)	<0.0001
Prescription	$260	(390)	$290	(395)	$225	(381)	<0.0001
(monthly) T2DM-related costs (2013 USD), mean (SD)
Total	$146	(220)	$168	(141)	$120	(285)	<0.0001
Medical	$54	(200)	$48	(110)	$61	(270)	<0.0001
Hospitalization	$3	(137)	$2	(68)	$5	(188)	0.0226
ER	$2	(31)	$1	(30)	$2	(32)	<0.0001
Physician office	$31	(69)	$29	(37)	$33	(93)	<0.0001
Laboratory	$5	(14)	$4	(14)	$5	(15)	0.0002
Radiology	$0	(7)	$0	(4)	$0	(10)	0.1319
Other outpatient	$14	(92)	$12	(59)	$16	(119)	<0.0001
Prescription	$92	(95)	$119	(90)	$59	(90)	<0.0001

DTMPR, dual therapy medication possession ratio; ER, emergency room; FDC, fixed-dose combination; LDC, loose-dose combination; MPR, medication possession ratio; PY, person years; SD standard deviation; T2DM, type 2 diabetes mellitus; USD, United States Dollars.

*p-values were obtained using t-tests for continuous variables and Chi square tests for categorical variables. p-values for differences in costs were obtained from the non-parametric Wilcoxon-Mann-Whitney test.

Adjusted outcomes

Data derived from the adjusted analysis are presented in Table 3. For the primary end-point of persistence, FDC therapy yielded significant reductions in non-persistence compared with LDC therapy (odds ratio = 0.79; 95% CI = 0.74–0.85; p < 0.001; hazard ratio = 0.88; 95% CI = 0.84–0.91; p < 0.001; Figure 2). Differences in most secondary outcomes were statistically significant in favor of FDC therapy (exceptions being the odds of hypoglycemia, the rate of ER visits, and all-cause medical costs), for which there were no differences between the two groups. Total monthly costs were lower for FDC vs LDC therapy, both all-cause ($1008 vs 1053; p = 0.006) and T2DM-related costs ($142 vs $155; p < 0.001).

Figure 2. Kaplan-Meier estimates of non-persistence. FDC, fixed-dose combination; LDC, loose-dose combination.

Table 3. Adjusted outcomes.

Characteristic	Parameter	FDC (n = 12,590)			LDC (n = 10,771)			p
		Estimate	95% CI
Odds of non-persistence (≥30-day gap)	OR*	0.79	[0.74, 0.85]		Reference group			<0.001
Hazard of non-persistence (≥30-day gap)	HR†	0.88	[0.84, 0.91]		Reference group			<0.001
Odds of adherence (MPR/DTMPR ≥0.8)	OR*	1.28	[1.20, 1.36]		Reference group			<0.001
Odds of hypoglycemic event occurrence	OR‡	0.80	[0.57, 1.14]		Reference group			0.213
Resource utilization
All-cause
Odds of hospitalization	OR‡	0.80	[0.72, 0.90]		Reference group			<0.001
Odds of ER visit	OR‡	0.76	[0.70, 0.82]		Reference group			<0.001
Relative difference in rate of visits
ER visits	IRR§	0.87	[0.74, 1.03]		Reference group			0.112
Physician office visits	IRR§	0.93	[0.91, 0.95]		Reference group			<0.001
Laboratory visits	IRR§	0.89	[0.85, 0.93]		Reference group			<0.001
Radiology visits	IRR§	0.89	[0.82, 0.97]		Reference group			0.007
Other outpatient visits	IRR§	0.90	[0.87, 0.95]		Reference group			<0.001
T2DM-related
Odds of hospitalization	OR*	0.74	[0.37, 1.45]		Reference group			0.376
Odds of ER visit	OR*	0.67	[0.53, 0.84]		Reference group			<0.001
Costs
(monthly) All-cause costs (2013 USD)¶		Predicted cost	95% CI Lower Limit	95% CI Upper Limit	Predicted cost	95% CI Lower Limit	95% CI Upper Limit
Total		$1008	$487	$3372	$1053	$524	$3416	0.006
Medical		$921	$296	$17,173	$957	$320	$17,145	0.060
Prescription (adjusting for index drug class [TZD, SU, and DPP4I])		$256	$152	$446	$282	$171	$477	<0.001
Prescription (not adjusting for index drug class)		$280	$272	$289	$237	$230	$244	<0.001
(monthly) T2DM-related costs (2013 USD)¶
Total		$142	$136	$149	$155	$149	$161	<0.001
Medical		$54	$31	$110	$58	$35	$117	<0.001
Prescription (adjusting for index drug class [TZD, SU, and DPP4I])		$156	$97	$258	$171	$109	$277	<0.001
Prescription (not adjusting for index drug class)		$119	$114	$125	$60	$58	$63	<0.001

CI, confidence interval; DPP4I, dipeptidyl peptidase-4 inhibitor; DTMPR, dual therapy medication possession ratio; ER, emergency room; FDC, fixed-dose combination; HR, hazard ratio; IRR, incidence rate ratio; LDC, loose-dose combination; MPR, medication possession ratio; OR, odds ratio; SD, standard deviation; SU, sulfonylurea; T2DM, type 2 diabetes mellitus; TZD, thiazolidinedione; USD, United States Dollars.

*OR obtained from logistic regression controlling for baseline demographic and clinical characteristics, index medication classes (DPP4I, SU, TZD), pre-index all-cause resource utilization, and pre-index prescription and total costs.

†HR obtained from Cox proportional hazards model controlling for baseline demographic and clinical characteristics, index medication classes (DPP4I, SU, TZD), pre-index all-cause resource utilization, and pre-index prescription and total costs.

‡OR obtained from logistic regression model controlling for baseline demographic and clinical characteristics, index medication classes (DPP4I, SU, TZD), and pre-index all-cause resource utilization.

§IRR obtained from zero-inflated negative binomial model controlling for baseline demographic and clinical characteristics, index medication classes (DPP4I, SU, TZD), and pre-index all-cause resource utilization.

¶Predicted costs obtained from statistical model—GLM controlling for baseline demographic and clinical characteristics, index medication classes (DPP4I, SU, TZD) (unless otherwise specified), pre-index all-cause resource utilization, and pre-index costs.

Each cost component is estimated from separate regression models and represents a predicted, rather than an observed value. Therefore, pharmacy and medical cost components may not add up to the total.

Discussion

Using data from 2009–2013, we characterized the relative patterns of use and health economic outcomes for FDC vs LDC in patients with treatment-naïve T2DM, with the findings collectively supporting the use of FDC in terms of persistence, adherence, hypoglycemia risk, resource utilization, and total and T2DM-related healthcare costs. The study is an important addition to the published research in this area, given its sound methodology, including the use of a single data source to capture treatment patterns as well as all the end-points leading to comprehensive results. With a person-years approach used for analysis (instead of the intent-to-treat approach), only outcomes that had occurred while the patient was still on treatment were associated with the treatment.

Our study used definitions (e.g., FDC, LDC, and MPR) consistent with prior studies, thus enabling comparison of results with prior findings. In the aforementioned study by Williams et al.5, a retrospective claims-based analysis based on prescription fills in 2007–2008, mean annual costs were likewise significantly lower with FDC vs LDC therapy both for all-cause ($8445 vs $9688; p = 0.004) and T2DM-related claims ($1641 vs $1900; p = 0.012); that study found a non-significant increase in T2DM-related pharmacy costs ($979 for FDC vs $964 for LDC; p = 0.506), but a significant reduction in medical costs ($663 for FDC vs $936 for LDC; p = 0.006). That study also found significant reductions in all-cause and T2DM-related inpatient stays and ambulatory visits, along with a significant reduction in all-cause ER visits (but not T2DM-related ER visits). In another of the more recent studies in this area, an analysis of patients enrolled in the ARNO Observatory (Italian) database in 2008, FDC was associated with higher rates of adherence vs LDC and was associated with lower mean total annual costs than monotherapy or dual therapy, which was attributed to lower mean costs for insulin and other drugs as well as diagnostic and specialist services that offset the higher prescription cost for FDC therapy18. From a clinical relevance standpoint, it is noteworthy that our study likewise compared adherence and persistence to FDC and LDC as therapy regimens as opposed to comparing individual drugs within these classes. Just over half of the FDC cohort was being treated with a DPP4 inhibitor/biguanide combination, with DPP4 inhibitors offering some benefits over sulfonylureas with respect to their impact on beta cell function and body weight19. Overall, adherence to OAD may have both clinical and economic impacts20. With many prior studies finding an association between OAD adherence and A1c, outcomes such as improved adherence are regarded as being suggestive of improved glycemic control, as was supported by a recent review that captured data across 23 studies21. Of note, direct assessment of glycemic control (using A1c data in a sub-group of patients) was of interest for our study; however, a paucity of available patients precluded this analysis.

Despite its strengths, this study is not without limitations. While claims data are an excellent source for understanding real-world treatment, there are some inherent limitations to this type of data source. Although cohort classification was based on prescription claims data, a prescription filled does not necessarily equate to treatment administration (at the same time, however, the study design required ≥2 prescription claims for classification, reducing the possibility of the drug not being administered). Of note, this may have over-estimated adherence; however, adherence differences between cohorts would not have been affected. It is appreciated that healthcare claims afford a large sample size and evaluation of real-world use, yet the lack of clinical data limits assessment of diabetes severity and other characteristics and risk factors that might influence physicians to initiate FDC vs LDC therapy. Although the analyses adjusted for covariates using proxies of severity from the available data source, some residual confounding may have been present, with a potential impact of differences in health status (rather than index therapies) on the findings. Importantly, diabetes-related medical costs were defined as claims with a primary diagnosis for diabetes—not including costs of claims due to chronic microvascular and macrovascular complications of diabetes, thereby under-estimating the costs incurred due to diabetes. Finally, while the results are generalizable to therapy-naïve patients with T2DM insured through employer-sponsored plans, they may not be applicable to other populations such as Medicaid.

In conclusion, early use of FDC therapy in treatment-naïve patients with T2DM has the potential to confer clinical benefits over those attainable with LDC therapy, not only in terms of persistence and adherence, but also resource utilization and costs of care.

Transparency

Declaration of funding

This study was funded by AstraZeneca.

Declaration of financial/other relationships

TL and WCL are currently employed at Xcenda, AmerisourceBergen Consulting Services. NS and JM are employed at Bristol-Myers Squibb (BMS); JM also owns BMS’s stocks. CS and ES are employed at AstraZeneca.

Acknowledgments

The authors would like to acknowledge Dr Pamela Landsman-Blumberg from Xcenda, AmerisourceBergen Consulting Services for providing her insight and expertise while designing the study.