Distribution and drivers of costs in type 2 diabetes mellitus treated with oral hypoglycemic agents: a retrospective claims data analysis

Abstract

Objective:

To describe the distribution of costs and to identify the drivers of high costs among adult patients with type 2 diabetes mellitus (T2DM) receiving oral hypoglycemic agents.

Methods:

T2DM patients using oral hypoglycemic agents and having HbA1c test data were identified from the Truven MarketScan databases of Commercial and Medicare Supplemental insurance claims (2004–2010). All-cause and diabetes-related annual direct healthcare costs were measured and reported by cost components. The 25% most costly patients in the study sample were defined as high-cost patients. Drivers of high costs were identified in multivariate logistic regressions.

Results:

Total 1-year all-cause costs for the 4104 study patients were $55,599,311 (mean cost per patient = $13,548). Diabetes-related costs accounted for 33.8% of all-cause costs (mean cost per patient = $4583). Medical service costs accounted for the majority of all-cause and diabetes-related total costs (63.7% and 59.5%, respectively), with a minority of patients incurring >80% of these costs (23.5% and 14.7%, respectively). Within the medical claims, inpatient admission for diabetes-complications was the strongest cost driver for both all-cause (OR = 13.5, 95% CI = 8.1–23.6) and diabetes-related costs (OR = 9.7, 95% CI = 6.3–15.1), with macrovascular complications accounting for most inpatient admissions. Other cost drivers included heavier hypoglycemic agent use, diabetes complications, and chronic diseases.

Limitations:

The study reports a conservative estimate for the relative share of diabetes-related costs relative to total cost. The findings of this study apply mainly to T2DM patients under 65 years of age.

Conclusions:

Among the T2DM patients receiving oral hypoglycemic agents, 23.5% of patients incurred 80% of the all-cause healthcare costs, with these costs being driven by inpatient admissions, complications of diabetes, and chronic diseases. Interventions targeting inpatient admissions and/or complications of diabetes may contribute to the decrease of the diabetes economic burden.

Keywords::

• Type 2 diabetes
• Cost drivers
• Cost distribution

Introduction

The number of Americans diagnosed with diabetes has increased remarkably over the past decade1–4. Globally, there are 347 million people living with diabetes5, with type 2 diabetes mellitus (T2DM) representing ∼95% of the adult cases4. Diabetes is associated with high morbidity and mortality: it is the leading cause of kidney failure, non-traumatic lower limb amputations, and acquired blindness6, and it doubles the patients’ risk of death from heart disease and stroke6.

The rising prevalence of diabetes and the changing healthcare practices have turned diabetes into a major contributor to the growing healthcare expenditure challenge: the total estimated cost of diabetes has increased by 21% in 5 years, from $174 billion in 2007 (or $202 billion in 2012 prices) to $245 billion in 20122,3. Although the overall economic burden of diabetes is growing in the US, the per capita diabetes expenditure adjusted for inflation has not changed much2,3: a diabetes patient was estimated to incur direct average annual healthcare costs of $13,741 in 2012, of which $7888 (57.4%) were diabetes-specific costs. However, with 10% of diabetes patients accounting for 68% of the healthcare expenditures7, mean cost estimates per patient have somewhat limited usefulness.

Previously, poor health status and older age were found to be associated with higher costs of treatment among newly diagnosed T2DM patients initiating oral hypoglycemic agents (OHAs)7, while diabetes complications (e.g., hypoglycemia and macrovascular complications) and certain chronic-state comorbidities (particularly heart disease, neurologic disorders, and renal disorders) were found to be more common among high-cost patients8 or to be associated with higher mean costs9–11. However, markers of diabetes severity—such as HbA1c and lipid levels—did not appear to affect costs when compared across cost-based patient quartiles8. While these were promising leads into finding the cost drivers of diabetes, many gaps in knowledge remain because prior studies either have considered only a limited number of potential cost drivers8,10,11, or were performed before the era of molecularly targeted drugs9.

The aim of the current study is to comprehensively evaluate and describe all-cause and diabetes-related costs incurred by T2DM patients treated with OHAs (with or without insulin), and to identify the main drivers of high costs in these patients.

Patients and methods

Data sources

The study used data from the 2004–2011 Truven Health Analytics MarketScan databases that include Commercial and Medicare Supplemental claims for ∼25 million individuals annually, covered by >130 health plans and self-insured employers. These databases contain information on patient demographics, enrollment history, inpatient (IP) and outpatient (OP) medical services, and pharmacy-related claims. For the sub-set of >1 million lives covered by health plans that record and report laboratory test information, laboratory test results were also available (Lab database). All census regions were represented, although the South and North Central (Midwest) regions were over-represented. Only Commercial and Medicare Supplemental claims data were used for the main study analysis. Data were de-identified and comply with HIPAA regulations12.

Sample selection and study design

The study sample included adult patients with T2DM receiving OHAs, identified using ICD-9 CM diagnostic codes for T2DM (250.x0 and 250.x2) on a minimum of two separate dates and NDC drug codes for OHAs.

The study used a cross-sectional design, with both cost outcomes and cost drivers measured over a fixed 1-year period (study period) following a patient-specific index date. To obtain a diverse population, index dates were randomly selected from the patient-specific sub-sets of OHA prescription dates satisfying the following criteria: (1) continuous health plan eligibility for a minimum 12 months before and minimum 12 months after the OHA prescription; (2) no insurance/coverage under a capitated or partially-capitated plan (HMO, POS) during the required continuous eligibility period; (3) ≥18 years of age at the time of the OHA prescription; and (4) ≥2 HbA1c measures during the 6 months before the OHA prescription and ≥2 HbA1c measures during the 6 months following the OHA prescription, to allow the estimation of duration of time within glycemic goal (Supplementary eFigure 1).

Patients were divided into two study cohorts—high vs lower cost—using the distribution of the total all-cause cost for the all-cause cost analyses and the distribution of the total diabetes-related costs for the diabetes-related cost analyses. In both analyses, the third quartile of the corresponding cost distribution was chosen as the cut-off point for defining the high and lower cost cohorts, so that the 25% most costly patients formed the high cost cohorts, while the 75% least costly patients formed the lower cost cohorts3,13.

Measurements

Direct healthcare costs were reported from a payer perspective (i.e., the amount reimbursed by the private insurer) over the 1-year study period: all-cause costs included all medical and pharmacy costs incurred during this period, while diabetes-related costs included only the sub-set of all-cause costs that were directly attributable to diabetes (i.e., costs for services that were associated with a T2DM diagnosis or with hypoglycemic agent, a definition that involves fewer assumptions than alternative definitions that also include costs associated with the diabetes-related complications and/or a percentage of the costs for illnesses unrelated to diabetes3). To be included in the diabetes category, an ICD-9 code for a T2DM diagnosis or an NDC code for an oral hypoglycemic agent were required to be listed on the claim; these codes could be located anywhere inside all populated medical claims fields. All-cause and diabetes-related costs were calculated for specific cost categories and sub-categories, including: medical service costs, which consisted of IP, OP, emergency room (ER), and other medical service costs (lab, radiology, or other ancillary services), and pharmacy costs, which consisted of hypoglycemic agent and other medication costs. Total costs comprised the sum of all medical and pharmacy cost components. Costs were inflated to 2011 prices using the Consumer Price Index-All Urban Consumers for Medical Care data (not seasonally adjusted, base period: 1982–1984 = 100), which is published by the Bureau of Labor Statistics of the US Department of Labor14.

Patients’ characteristics were measured during the 1-year study period, index date included. Demographic data included age, gender, region of residence, and type of health plan at the index date. The use of hypoglycemic agents during the study period was measured by the type and number of classes of hypoglycemic agents used15, use of bile acid sequestrates, presence of treatment gaps exceeding 90 consecutive days, and proportion of days covered (PDC) by an hypoglycemic agent. The drug treatment classes used in this analysis were: alpha-glucosidase inhibitors, amylin analogs, hypoglycemic agent combinations, biguanide sensitizers, dipeptidyl peptidase-4 (DPP-4) inhibitors, human insulin, incretin mimetic agents (GLP-1 receptor agonists), meglitinide analogs, sulfonylureas, thiazolidinediones (TZDs), and others. For descriptive purposes, the type of drug and the year of the index OHA prescription were also measured. Patients were classified as being within the glycemic goal at the index date if their most recent HbA1c lab test result showed blood levels <7% NGSP (i.e., <53 mmol/mol IFCC)16. Patients’ average HbA1c levels during the study period and the percentage of time they stayed within the glucose goal were assessed by the last value carried forward method. Complications of diabetes and general/diabetes-related comorbidities were identified from the ICD9 diagnoses recorded during the study period17. Complications of diabetes17 consisted of microvascular complications (including diabetic retinopathy, nephropathy, and neuropathy), macrovascular complications (including atherosclerosis, aneurysm, or embolism, peripheral vascular conditions, cerebrovascular disease, and coronary artery disease), and non-vascular complications, such as infections (urinary tract, kidney, and mycoses), ocular complications (other than diabetic retinopathy), ulceration or diabetic bone changes, ketoacidosis or hyperosmolarity, hypoglycemia, diabetic coma, and unspecified complications of diabetes. Healthcare resource utilization during the study period was measured by the frequency of patients with a visit to an endocrinologist or with an IP admission or ER visit related to a complication of diabetes. Patients’ overall comorbidity burden during the study period was measured by the Deyo-Charlson comorbidity index18 (extracted from all claims, including inpatient, outpatient, emergency department, and other medical visits), modified to exclude diabetes.

Statistical analyses

For all costs, the following distributional statistics were reported: mean (± standard deviation, SD) and median (interquartile range [IQR]) costs per patient, relative share of each cost component to the total cost, proportion of patients incurring costs in each cost component, and the proportion of patients incurring 80% of costs in each cost component (starting from the patient with highest cost). Patient characteristics were compared between the high and the lower cost cohorts using t-tests and chi-square tests, respectively.

The drivers of all-cause and diabetes-related high costs were identified in multivariate logistic regression models with a binary indicator of high vs lower cost cohort as dependent variable and patient characteristics as potential cost drivers, and reported as odds ratios (OR) and 95% confidence intervals (CIs).

In addition, exploratory post-hoc analyses were conducted to: (1) identify specific complications of diabetes that accounted for the strong effect that IP admissions related to a complication of diabetes had on the patient’s risk of incurring high all-cause and diabetes-related costs; and (2) explore differences in the effect of sex on all-cause vs diabetes-related high cost.

In all analyses, statistical significance was based on a two-sided p-value of 0.05. Analyses were performed using SAS v.9.3 (Cary, NC) software.

Results

Description of the study sample

A total of 4104 T2DM patients met the study selection criteria (Supplementary eFigure 1). The mean patient age was 53.9 years (SD = 8.4 years), 3% of patients were aged ≥65 years, and 42.5% of the patients were female. Per the sample selection criteria, all patients had their HbA1c levels measured before the index date; a total of 1950 patients (47.5%) had HbA1c values <7% at the index date. Only 71 patients (1.7%) were new users of OHAs (Supplementary eTable 1).

Distribution of all-cause and diabetes-related costs

The total 1-year all-cause cost, summed up for the 4104 T2DM study patients, was $55,599,311. Of the $55,599,311 total cost, 33.8% was accounted for by diabetes-related costs. All-cause costs were comprised of 63.7% medical service and 36.3% pharmacy costs, with the majority of the medical service costs being due to OP (53.4%) and IP (36.3%) costs (Table 1). Diabetes-related costs accounted for 33.8% of the total costs and were comprised of medical service (59.5%) and hypoglycemic agent pharmacy (40.5%) costs. OP (42.2%) and IP (50.2%) were the most costly components of diabetes-related medical service costs.

Table 1. Distribution of all-cause and diabetes-related costs among type 2 diabetes patients with oral hypoglycemic agent and lab test usage.

	Distribution of costs across the 4104 patients in the study sample				Patients incurring 80% of the costs*
Cost components	Share of total costs, %	Share of patients incurring costs, %	Mean cost (SD), $	Median cost (IQR), $	Number of patients, n	Share of all patients, %
All-cause costs^†
Pharmacy	36.3	100	4913 (5042)	3822 (2024–6372)	2071	50.5
Hypoglycemic agent	13.7	100	1855 (1922)	1469 (223–2751)	1658	40.4
Other pharmacy	22.6	99	3058 (4387)	2082 (937–3767)	1816	44.2
Medical	63.7	100	8635 (21670)	2431 (1016–6933)	963	23.5
IP	23.1	11	3132 (14990)	0 (0–0)	190	4.6
ER	1.4	15	195 (927)	0 (0–0)	256	6.2
OP	34.0	99	4610 (11133)	1953 (890–4599)	1347	32.8
Other medical services	5.2	51	698 (5073)	24 (0–338)	454	11.1
Total	–	100	13,548 (23260)	7389 (4188–13714)	1728	42.1
Diabetes-related costs^‡
Pharmacy (hypoglycemic agent)	40.5	100	1855 (1922)	1469 (223–2751)	1658	40.4
Medical	59.5	99	2728 (10187)	620 (384–1149)	604	14.7
IP	29.9	7	1370 (8873)	0 (0–0)	112	2.76
ER	1.7	7	76 (430)	0 (0–0)	127	3.1
OP	25.1	97	1151 (3404)	530 (310–874)	1481	36.1
Other medical services	2.9	35	131 (658)	0 (0–68)	330	8.0
Total	–	100	4583 (10449)	2546 (961–4361)	1558	38.0

ER, emergency room; IQR, interquartile range from Q1–Q3; IP, inpatient; OP, outpatient; SD, standard deviation.

*Costs incurred by patients were added in descending order (i.e., starting from the patient with highest costs) until the 80% cost cut-off was reached.

†All-cause costs were defined as the sum of all costs for medical services and medications, regardless of diagnosis or type of medication.

‡Diabetes-related costs were defined as the sum of claims for a medical service associated with a diagnosis for diabetes (i.e., ICD-9 CM codes 250.x) or claims for a hypoglycemic agent.

While virtually all patients (>99%) incurred some all-cause and diabetes-related pharmacy and medical service costs, not all contributed equally to the share of costs: 80% of medical service costs were accounted for by the 23.5% and 14.7% patients with the highest all-cause and diabetes-related medical service costs, respectively (Table 1). The trend was similar when examining OP costs. In contrast, few patients incurred all-cause or diabetes-related IP costs (11% and 7%, respectively), but their costs weighed heavily on the corresponding total cost (they accounted for 23.1% and 29.9%, respectively, of the total costs and 36.3% and 50.2%, respectively, of the medical costs).

Based on these cost distributions, the cut-offs used to define the high cost cohorts were $13,714 for all-cause costs and $4361 for diabetes related costs (fourth quartile, Table 1).

Comparison of patient characteristics between high vs lower cost cohorts

When comparing the high vs lower all-cause cost cohorts, the high-cost cohort included slightly older patients, more females (46.8% vs 41.1%), more patients with medical encounters for both diabetes complications and for general and type 2 diabetes-associated comorbidities, and more patients with visits to the ER or endocrinologist during the study period (all p < 0.01) (Table 2). The high-cost cohort also included patients who had higher use of hypoglycemic agent combinations at the index date, and higher use of insulin, incretin mimetic agents, and multiple classes of hypoglycemic agents during the study period.

Table 2. Characteristics of study population: comparison high all-cause cost cohort vs lower all-cause cost cohort.

Characteristics	High cost cohort (n = 1026)	Lower cost cohort (n = 3078)	p-Value*
Demographics at index date
Age, Mean (SD)	55.3 (8.0)	53.4 (8.5)	<0.001†
Female, n (%)	480 (46.8)	1264 (41.1)	0.001†
Region, n (%)
Northeast	129 (12.6)	349 (11.3)	0.286
North Central	236 (23.0)	593 (19.3)	0.010†
South	552 (53.8)	1841 (59.8)	<0.001†
West	109 (10.6)	293 (9.5)	0.303
Insurance plan type, n (%)
Comprehensive	64 (6.2)	134 (4.4)	0.015†
Non-Capitated Point-of-Service (POS)	200 (19.5)	652 (21.2)	0.248
Preferred Provider Organization (PPO)	710 (69.2)	2160 (70.2)	0.555
Consumer-Driven Health Plan (CDHP)	39 (3.8)	102 (3.3)	0.458
High Deductible Health Plan (HDHP)	10 (1.0)	20 (0.6)	0.290
Hypoglycemic agents used during the 1-year study period
Hypoglycemic Agents, n (%)
Alpha-Glucosidase Inhibitors	11 (1.1)	27 (0.9)	0.572
Hypoglycemic Agents - Amylin Analogs	7 (0.7)	6 (0.2)	0.016†
Hypoglycemic Agent Combinations	225 (21.9)	818 (26.6)	0.003†
Biguanide Sensitizers	681 (66.4)	2126 (69.1)	0.108
Dipeptidyl Peptidase-4 (DPP-4) Inhibitors	198 (19.3)	505 (16.4)	0.033†
Human Insulin	420 (40.9)	562 (18.3)	<0.001†
Incretin Mimetic Agents (GLP-1 Receptor Agonists)	170 (16.6)	260 (8.4)	<0.001†
Meglitinide Analogues	22 (2.1)	50 (1.6)	0.272
Sulfonylureas	440 (42.9)	1216 (39.5)	0.056
Thiazolidinediones (TZDs)	318 (31.0)	900 (29.2)	0.287
Other	38 (3.7)	46 (1.5)	<0.001†
Number of different hypoglycemic drug classes, Mean (SD)	3.0 (1.3)	1.8 (0.9)	<0.001†
At least 2 different hypoglycemic drug classes, n (%)	905 (88.2)	581 (56.6)	<0.001†
At least 3 different hypoglycemic drug classes, n (%)	631 (61.5)	200 (19.5)	<0.001†
Use of hypoglycemic agents
Treatment gaps ≥90 consecutive days, n (%)	113 (11.0)	387 (12.6)	0.186
Proportion of days of use during the period, Mean (SD)	85.2 (20.0)	82.7 (20.9)	<0.001†
Used hypoglycemic agents for more than 80% of the period, n (%)	771 (75.1)	2173 (70.6)	0.005†
Use of bile acid sequestrants	32 (3.1)	47 (1.5)	<0.001†
HbA1c lab test results
Within glycemic goal (i.e., HbA1c level <7% NGSP [<53 mmol/mol IFFC]) at the index date, n (%)	479 (46.7)	1471 (47.8)	0.539
Average HbA1c level during the 1-year study period, Mean (SD)
In % NGSP units	7.3 (1.3)	7.2 (1.2)	0.354
In mmol/mol IFCC units	56 (14.2)	55 (13.1)
Percentage of time within glycemic goal during the 1-year study period (i.e., HbA1c level <7%), Mean (SD)	49.7 (43.2)	50.6 (43.1)	0.464
Complications of diabetes identified during the 1-year study period
Microvascular complications
Diabetic retinopathy	161 (15.7)	292 (9.5)	<0.001†
Nephropathy	173 (16.9)	206 (6.7)	<0.001†
Neuropathy	239 (23.3)	279 (9.1)	<0.001†
Macrovascular complications
Atherosclerosis, aneurysm, or embolism	70 (6.8)	63 (2.0)	<0.001†
Peripheral vascular conditions	150 (14.6)	153 (5.0)	<0.001†
Cerebrovascular disease	113 (11.0)	87 (2.8)	<0.001†
Coronary artery disease	305 (29.7)	235 (7.6)	<0.001†
Non-vascular complications
Infections	241 (23.5)	386 (12.5)	<0.001†
Other ocular complications	247 (24.1)	452 (14.7)	<0.001†
Ulceration or diabetic bone changes	49 (4.8)	31 (1.0)	<0.001†
Ketoacidosis or hyperosmolarity	12 (1.2)	12 (0.4)	0.005†
Hypoglycemia	70 (6.8)	48 (1.6)	<0.001†
Diabetic coma	7 (0.7)	25 (0.8)	0.682
Unspecified diabetes-related complication	78 (7.6)	120 (3.9)	<0.001†
Health resource utilization during the 1-year study period
IP admission related to a complication, n (%)	208 (20.3)	20 (0.6)	<0.001†
ER visit related to a complication, n (%)	49 (4.8)	30 (1.0)	<0.001†
Patients seen by an endocrinologist, n (%)	195 (19.0)	353 (11.5)	<0.001†
Comorbidity profile measured during the 1-year study period
Charlson comorbidity index (without diabetes), Mean (SD)	1.3 (1.7)	0.3 (0.7)	<0.001†
Comorbidity associated with diabetes
Hypertension and associated conditions	679 (66.2)	1709 (55.5)	<0.001†
Lipid disorders	723 (70.5)	2106 (68.4)	0.220
Congestive heart failure	85 (8.3)	34 (1.1)	<0.001†
Overweight, obesity	109 (10.6)	151 (4.9)	<0.001†
Mental disorder	117 (11.4)	109 (3.5)	<0.001†
Other genitourinary system disorder	248 (24.2)	301 (9.8)	<0.001†
Other (physical or mental) comorbid conditions
AIDS	10 (1.0)	0 (0.0)	–
Anemia	184 (17.9)	180 (5.8)	<0.001†
Chronic pulmonary disease	154 (15.0)	200 (6.5)	<0.001†
Coagulopathy	36 (3.5)	20 (0.6)	<0.001†
Fluid electrolyte disorders	98 (9.6)	59 (1.9)	<0.001†
Hypothyroidism	189 (18.4)	336 (10.9)	<0.001†
Liver disease	76 (7.4)	43 (1.4)	<0.001†
Cancer	134 (13.1)	88 (2.9)	<0.001†
Other neurological disorders	58 (5.7)	27 (0.9)	<0.001†
Paralysis	6 (0.6)	3 (0.1)	0.004†
Psychoses	90 (8.8)	60 (1.9)	<0.001†
Pulmonary circulation disorders	27 (2.6)	9 (0.3)	<0.001†
Rheumatoid arthritis	68 (6.6)	41 (1.3)	<0.001†
Valvular disease	50 (4.9)	61 (2.0)	<0.001†
Behavioral factors
Alcohol abuse	2 (0.2)	4 (0.1)	0.637
Drug abuse	4 (0.4)	5 (0.2)	0.177
Smoker	16 (1.6)	24 (0.8)	0.028†
Weight loss	20 (1.9)	14 (0.5)	<0.001†

AIDS, acquired immune deficiency syndrome; ER, emergency room; HbA1c, glycated hemoglobin; IFFC, International Federation of Clinical Chemistry; IP, inpatient; n, number; NGSP, National Glycohemoglobin Standardization Program; SD, standard deviation.

*Comparing high and lower cost cohorts using chi-square or Fisher tests for categorical variables and t-tests for continuous variables.

†Differences between the high and lower cost cohorts were statistically significant at p < 0.05.

Similar patient characteristics differences were noted between the cohorts defined based on the diabetes-related costs (data not shown), except that there were fewer female than male patients in the high-cost cohort (39.0% vs 43.7%, p < 0.05), and the differences related to drug use and lab tests were more visible in the diabetes-related than all-cause cost analyses. In particular, when compared to the lower cost cohort, the high-cost cohort in the diabetes-related analyses had higher use of hypoglycemic agents (all classes, except biguanide insulin sensitizers), and had poorer glycemic control (34.3% patients in the high cost cohort were within the glycemic goal at index date vs 51.9% in the lower cost cohort, while the mean HbA1c during the study period was 7.6% NGSP [60 mmol/mol IFFC] in the high cost cohort vs 7.1% NGSP [54 mmol/mol IFFC] in the lower cost cohort). In addition, patients in the diabetes-related high-cost cohort were more likely to have had ER visits for a complication of diabetes during the study period than subjects in the lower cost cohort (data not shown).

High-cost drivers

Cost driver analyses showed that complications of diabetes and comorbidities were the most important factors leading to high all-cause costs (Table 3). The five most influential high-cost drivers of all-cause costs were IP admissions for a complication of diabetes (OR = 13.5), liver diseases (OR = 5.3), cancer (OR = 5.2), rheumatoid arthritis (OR = 4.5), and neurological disorders (OR = 3.0). For diabetes-related costs, the most important factors leading to high costs were complications of diabetes, insulin use, and comorbidities, while the five most influential high cost drivers were IP admissions for a complication of diabetes (OR = 9.7), insulin use (OR = 3.1), ER visit for a complication of diabetes (OR = 2.3), endocrinologist visit (OR = 1.9), and obesity (OR = 1.7)

Table 3. All-cause and diabetes-related high cost drivers among type 2 diabetes patients receiving OHAs.

Potential cost drivers*	All-cause costs		Diabetes-related costs
	OR (95% CI)	p-Value	OR (95% CI)	p-Value
Demographics at index date
Age (per year increase)	1.00 (0.99;1.01)	0.772	0.99 (0.98;1.00)	0.026†
Female (vs Male)	1.06 (0.88;1.29)	0.518	0.81 (0.67;0.97)	0.025†
Region
Northeast	1.14 (0.85;1.52)	0.380	0.77 (0.56;1.03)	0.085
North Central	1.26 (0.98;1.60)	0.066	1.33 (1.05;1.68)	0.019†
South (Reference group)	1	–	1	–
West	1.23 (0.90;1.67)	0.189	0.92 (0.67;1.25)	0.582
Insurance Plan Type
Non-Capitated Point-of-Service (POS)	0.81 (0.56;1.17)	0.259	0.74 (0.51;1.09)	0.125
Preferred Provider Organization (PPO)	0.78 (0.57;1.08)	0.128	1.15 (0.83;1.61)	0.397
Other plans^‡ (Reference group)	1	–	1	–
Hypoglycemic agent use during the 1-year study period
Hypoglycemic agents
Human Insulin	2.42 (1.93;3.03)	<0.001†	3.15 (2.59;3.85)	<0.001†
Number of hypoglycemic drug class used	1.16 (1.07;1.26)	<0.001†	1.68 (1.55;1.82)	<0.001†
Use of hypoglycemic agents
Proportion of days of use during the study period (per 1% increase)	1.01 (1.00;1.01)	0.026†	1.02 (1.02;1.03)	<0.001†
Use of bile acid sequestrants	1.29 (0.71;2.28)	0.386	0.99 (0.55;1.74)	0.970
Lab test information
Percentage of time within glycemic goal during the study period
<3 months	0.73 (0.57;0.94)	0.016†	1.05 (0.83;1.33)	0.673
3–9 months (Reference group)	1	–	1	–
≥9 months	1.11 (0.86;1.42)	0.430	0.93 (0.73;1.20)	0.573
Complications of diabetes identified during the 1-year study period
Microvascular complications
Diabetic retinopathy	1.06 (0.80;1.40)	0.667	1.17 (0.90;1.51)	0.237
Nephropathy	1.30 (0.96;1.73)	0.085	1.23 (0.92;1.64)	0.161
Neuropathy	1.72 (1.34;2.20)	<0.001†	1.50 (1.17;1.92)	0.001†
Macrovascular complications
Atherosclerosis, aneurysm, or embolism	2.01 (1.24;3.22)	0.004†	1.42 (0.86;2.32)	0.169
Peripheral vascular conditions	1.17 (0.83;1.64)	0.354	1.21 (0.87;1.67)	0.254
Cerebrovascular disease	1.48 (0.99;2.20)	0.055	0.74 (0.49;1.13)	0.171
Coronary artery disease	2.83 (2.19;3.65)	<0.001†	1.56 (1.20;2.03)	<0.001†
Non-vascular complications
Infections	0.89 (0.67;1.19)	0.443	0.93 (0.70;1.24)	0.638
Other ocular complications	1.34 (1.06;1.69)	0.012†	1.54 (1.23;1.93)	<0.001†
Ulceration or diabetic bone changes	1.25 (0.65;2.37)	0.497	1.83 (0.99;3.42)	0.056
Ketoacidosis or hyperosmolarity	1.91 (0.61;5.62)	0.252	0.88 (0.25;2.96)	0.842
Hypoglycemia	1.46 (0.84;2.51)	0.177	1.69 (1.01;2.85)	0.046†
Diabetic coma	0.37 (0.10;1.14)	0.112	0.52 (0.18;1.36)	0.201
Unspecified diabetes-related complication	1.25 (0.83;1.86)	0.275	1.08 (0.73;1.59)	0.682
Healthcare utilization during the 1-year study period
Any IP admission related to a complication	13.54 (8.14;23.57)	<0.001†	9.69 (6.31;15.08)	<0.001†
Any ER visit related to a complication	1.90 (1.01;3.59)	0.047†	2.33 (1.28;4.23)	0.006†
Patient seen by an endocrinologist	1.23 (0.96;1.59)	0.105	1.91 (1.51;2.42)	<0.001†
Comorbidity profile measured during the 1-year study period
Comorbidity associated with diabetes
Hypertension and associates condition	1.20 (1.00;1.45)	0.052	1.07 (0.89;1.28)	0.477
Lipid disorders	1.16 (0.95;1.42)	0.150	1.28 (1.05;1.57)	0.016†
Congestive heart failure	1.69 (0.97;2.96)	0.065	0.62 (0.37;1.03)	0.070
Overweight, obesity	1.76 (1.25;2.46)	0.001†	1.75 (1.24;2.46)	0.001†
Mental disorder	2.39 (1.55;3.66)	<0.001†	1.36 (0.86;2.13)	0.188
Other genitourinary system disorder	2.04 (1.52;2.74)	<0.001†	1.55 (1.15;2.09)	0.004†
Other (physical or mental) comorbid conditions
AIDS	–	–	0.42 (0.07;2.21)	0.317
Anemias (blood loss or deficiency)	1.61 (1.20;2.16)	0.002†	1.16 (0.85;1.56)	0.348
Chronic Pulmonary Disease	1.72 (1.28;2.32)	<0.001†	1.03 (0.75;1.40)	0.868
Coagulopathy	1.67 (0.79;3.53)	0.176	0.85 (0.40;1.74)	0.672
Fluid Electrolyte Disorders	2.59 (1.66;4.02)	<0.001†	1.56 (1.02;2.38)	0.039†
Hypothyroidism	1.49 (1.16;1.92)	0.002†	0.98 (0.75;1.28)	0.898
Liver Disease	5.35 (3.37;8.52)	<0.001†	1.43 (0.88;2.30)	0.147
Cancer (metastatic, solid tumor or lymphoma)	5.16 (3.69;7.23)	<0.001†	1.58 (1.09;2.26)	0.014†
Other Neurological Disorders	3.00 (1.64;5.53)	<0.001†	0.69 (0.37;1.27)	0.240
Paralysis	1.42 (0.19;10.04)	0.718	0.87 (0.15;4.70)	0.872
Psychoses	2.46 (1.46;4.14)	<0.001†	1.05 (0.60;1.83)	0.862
Pulmonary Circulation Disorders	1.48 (0.54;4.18)	0.447	1.01 (0.38;2.64)	0.978
Rheumatoid Arthritis	4.49 (2.76;7.34)	<0.001†	1.06 (0.62;1.78)	0.821
Valvular Disease	1.03 (0.60;1.73)	0.912	1.02 (0.60;1.73)	0.928

CI, confidence interval; ER, emergency room; IP, inpatient; OHA, oral hypoglycemic agents; OR, odds ratio (an OR >1 shows that the potential cost driver increases the patient risk of being in the high cost cohort).

*If two patient characteristics were highly correlated with one another, only one was kept in the model, in order to prevent co-linearity. The Deyo-Charlson comorbidity score was excluded as a potential factor in this analysis, in order to observe the individual effects of specific comorbidities. Given that there are only 10 study subjects with AIDS and that all these patients are in the high all-cause cohort, the effect of AIDS is not estimated in the all-cause cost model.

†The effect of the covariate on the odds of incurring high costs was statistically significant at p < 0.05.

‡Include comprehensive, CDHP, and HDHP plans, which together accounted for <10% of sample.

Because IP admission related to a complication was the strongest cost driver for both all-cause and diabetes-related costs, a post-hoc analysis where results were sorted by complication category (microvascular, macrovascular, and non-vascular) was performed. In these analyses, IP admissions for macrovascular complications were most prevalent, occurring in 70% of the patients with an IP admission related to complications (Table 4). These patients also had the highest costs ($30,366) in all-cause cost analyses and the highest likelihood of being in the high cost cohort (OR = 13.4). However, in diabetes-related cost analyses, patients with non-vascular complications had the highest costs ($15,640) and had the highest likelihood of being in the high cost cohort (OR = 16.2). When specific complication diagnoses were investigated, overall, IP admissions for coronary artery diseases were the most prevalent (47% of the patients with complication-related IP admissions). In the linear regression models, infections ($30,547) and coronary artery diseases ($26,803) were the most influential drivers of all-cause costs, while hypoglycemia ($29,278) and atherosclerosis ($16,148) were the most influential drivers of diabetes-related costs.

Table 4. Inpatient admission related to a complication of diabetes, breakdown by type of admission.

		Estimates from adjusted* logistic and linear regression models
Modeling of IP admission in three alternative models (any vs none)	n (%) (of 4104 patients in	All-cause costs		Diabetes-related costs
	study sample)	OR^† (95% CI)	Cost differences^‡, $ (95% CI)	OR^† (95% CI)	Cost differences^‡, $ (95% CI)
Model 1: main analysis
Inpatient admission related to a complication	228 (5.6)	13.54 (8.14;23.57)^§	+32,200 (29,275;35,125)^§	9.69 (6.31;15.08)^§	+15,012 (13,570;16,453)^§
Model 2: IP admission variable from main analysis was replaced by specific categories of IP admission
IP admission for microvascular complications	78 (1.9)	9.02 (2.87;40.36)^§	+18,801 (13,911;23,692)^§	1.75 (0.82;3.82)	+8,691 (6,293;11,088)^§
IP admission for macrovascular complications	159 (3.9)	13.36 (7.02;27.45)^§	+30,366 (26,871;33,861)^§	5.55 (3.35;9.30)^§	+11,466 (9,753;13,180)^§
IP admission for non-vascular complications	70 (1.7)	9.50 (3.32;31.76)^§	+18,900 (13,823;23,977)^§	16.24 (6.72;43.12)^§	+15,640 (13,151;18,129)^§
Model 3: IP admission variable from main analysis was replaced by specific diagnoses related to IP admission^‖
IP admission for nephropathy complications	53 (1.3)	4.96 (1.29;32.98)^§	+16,107 (10,106;22,108)^§	0.83 (0.35;2.05)	+3,241 (356;6,126)^§
IP admission for neuropathy complications	30 (0.7)	21.93 (3.84;418.93)^§	+12,856 (5,489;20,222)^§	25.23 (4.58;473.95)^§	+14,269 (10,727;17,810)^§
IP admission for atherosclerosis, aneurysm, or embolism complications	12 (0.3)	–	+23,035 (11,704;34,366)^§	3.24 (0.64;19.75)	+16,148 (10,700;21,595)^§
IP admission for peripheral vascular conditions complications	36 (0.9)	1.65 (0.47;6.81)	+21,725 (14,517;28,933)^§	0.91 (0.29;2.93)	+2,189 (−1,276;5,655)
IP admission for cerebrovascular disease complications	37 (0.9)	11.53 (3.08;57.17)^§	+12,817 (6,047;19,586)^§	4.72 (1.78;12.73)^§	+995 (−2,260;4,250)
IP admission for coronary artery disease complications	108 (2.6)	11.86 (5.53;28.66)^§	+26,803 (22,736;30,869)^§	4.83 (2.74;8.68)^§	+11,965 (10,009;13,920)^§
IP admission for infections (urinary tract infection, kidney infection, mycoses) complications	27 (0.7)	6.52 (1.40;35.89)^§	+30,547 (23,011;38,082)^§	11.16 (3.51;37.70)^§	+12,807 (9,184;16,430)^§
IP admission for ulceration or diabetic bone changes complications	12 (0.3)	–	+609 (−12,798;14,016)	0.85 (0.07;20.84)	−11,296 (−17,742;−4,850)^§
IP admission for hypoglycemia complications	22 (0.5)	–	+11,839 (1,751;21,928)^§	–	+29,278 (24,427;34,128)^§
IP admission for unspecified diabetes-related complication complications	19 (0.5)	14.79 (2.20;299.02)^§	+11,561 (2,407;20,716)*	30.30 (5.11;584.56)^§	+15,359 (10,958;19,761)^§

CI, confidence interval; IP, inpatient; n, number; OR, odds ratio.

*Models are adjusted for all covariates from the main model. In Models 2 and 3, the logistic regression model from the main analysis was replicated by replacing the IP admission related to a complication of diabetes by either specific categories of IP admissions (i.e., microvascular, macrovascular, and non-vascular complications [model 2]) or by the specific diagnoses associated with IP admissions (e.g., nephropathy, neuropathy, hypoglycemia [model 3]). Then, the logistic regression models 2 and 3 were replaced by linear models in which the dependent variable was a continuous measure of total cost per patient during the 1-year study period, while the independent variables remained the same as in models 2 and 3. The estimates yielded by the linear model represent the incremental costs associated with each specific complication of diabetes.

^†An OR >1 indicates that patients with any given type of admission were more likely to have high costs (i.e., quartile 4).

^‡Cost difference is an incremental annual total cost of patients with any given type of admission.

^§The complication of diabetes had a statistically significant effect on incurring high costs at p < 0.05.

^‖Diagnoses affecting few patients were excluded from some regression models to avoid non-convergence.

Sex-stratified analyses

Additional post-hoc analyses were performed to investigate the finding that females were over-represented in the all-cause high cost cohort, while males were over-represented in the diabetes-related high cost cohort. A comparison of the distribution of the all-cause and diabetes-related costs in males and females showed that the average costs were similar in males vs females. However, the distribution of costs between different cost components was different for males vs females: males had higher costs compared to females for hypoglycemic agents, while females had higher costs compared to males for all-cause medical service costs, especially for OP costs (Supplementary eTable 2).

When analyzing patient characteristics, it was noted that males and females in the study sample had similar ages (mean age = 54.0 in males vs 53.8 in females), but exhibited differences in other key comparators, such as drug use, glycemic control, complications of diabetes, and other comorbidities. Specifically male patients were more likely than female patients to use hypoglycemic agent combinations, sulfonylureas, and TZDs (27.4% vs 22.7%, 41.9% vs 38.2%, and 32.8% vs 25.5%, respectively (p < 0.05)); used slightly more hypoglycemic drug classes (mean 2.5 vs 2.4 in females, p < 0.05); exhibited poorer glycemic control (46.0% within glycemic goal at index date vs 49.6% for females, p < 0.05); were more likely to have certain complications of diabetes (e.g., atherosclerosis, aneurysm or embolism, 3.9% vs 2.4% in females, coronary artery diseases, 15.3% vs 10.3%, and ulceration and diabetic bone changes, 2.5% vs 1.2%, p < 0.05).

On the other hand, female patients were more likely to use biguanide sensitizers (71.2% vs 66.4% in males, p < 0.05), incretin (12.0% vs 9.4%, p < 0.05), and bile acid (2.5% vs 1.5%, p < 0.05); to have infection complications (22.5% vs 10.0% in males, p < 0.05); and to have diabetes-associated and other comorbidities, such as obesity, mental disorders, genitourinary disorders, anemia, chronic pulmonary disease, hypothyroidism, cancer, psychoses, and rheumatoid arthritis (data not shown).

Discussion

Achieving a measure of control over the high cost of diabetes would require a thorough understanding of the distribution of costs and the factors that drive high costs among diabetes patients. Previous literature has touched upon determinants of high diabetes expenditures at the patient level, but has not covered a comprehensive set of potential factors7,8,10,11, and may even be outdated9. Using a comprehensive list of covariates, the study found that complications of diabetes and comorbidities were the most important factors which led to high costs, with IP admissions for complications of diabetes, especially macrovascular complications, being the strongest driver of both all-cause and diabetes-related high costs. This study demonstrates that the economic burden of T2DM belongs disproportionally to a minority of patients.

The cost estimates presented in this study for T2DM patients are in line with the latest figures released by the American Diabetes Association (ADA) for 2012: average annual costs per patient $13,548 vs $13,7413 for all-cause costs and $4583 and $78883, respectively, for diabetes-related costs. However, these similarities observed in all-cause cost estimates may be coincidental, given that the distribution of age, a significant driver of diabetes costs3,7, was different between the two studies. Also, the higher diabetes-related cost observed in the ADA study is likely reflecting the fact that diabetes-related costs were restricted to costs directly attributable to diabetes in this study, while in the ADA study they also included costs attributable to diabetes-associated co-morbid conditions and some proportion of all other healthcare costs.

Prior studies have indicated that a minority of new patients (10%) incur the majority (68%) of the costs7. In this study, a similar trend was observed for medical all-cause and diabetes-related costs, where 23.5% and 14.7% of the patients, respectively, incurred 80% of the costs. The trend was even more dramatic for IP costs (11% and 7%, respectively, incurred all IP costs), highlighting the important economic benefit that can be achieved by improving disease prevention and treatment.

This study confirms previous findings indicating that IP stays are a major determinant of healthcare costs for T2DM patients9,19, and further clarifies that microvascular (neuropathy, in particular)20 and macrovascular complications8,11,19,21,22 and infections23 are correlated with higher costs of all-cause IP admissions. In terms of diabetes-related costs, we found that non-vascular complications, including hypoglycemia10, were correlated with the high costs of IP admissions. Among comorbidities contributing to the high costs of care experienced by T2DM patients, hypertension and mental and other neurological disorders9 have been previously identified as potential cost drivers for all-cause healthcare costs; to these we added genitourinary system disorders as a source of high costs in both all-cause and diabetes-related healthcare expenses.

The observation that male gender appears to be a driver of high diabetes-related costs may seem puzzling given that, in the all-cause cost analysis, there was no statistically significant gender difference (although, numerically, females were over-represented in the high all-cause healthcare costs group of T2DM patients). However, these findings may be explained by the fact that women tended to have a higher overall comorbidity burden (such as depression, which occurs more frequently in women24,25, and results in higher costs26), which affects all-cause costs, while men tended to be more susceptible to macrovascular (diabetes-related/diabetes-causing) conditions11,27, as well as to have more complex hypoglycemic agent treatments. Thus, these findings suggest that, when designing strategies to effectively manage the disease burden of diabetes, it is important to take into account the gender-associated differences in costs and comorbidity profiles of T2DM patients.

Some of the drivers found to be influential in this study occurred with relatively rare frequency in the study sample, such as hypoglycemia for T2DM costs or liver disease for all-cause costs. However, even though these conditions are not frequent in our study sample, they occurred more frequently in the high-cost than lower-cost cohort: 7.0% of the patients in the high-cost T2DM cohort had hypoglycemia vs 1.5% of the patients in the lower-cost T2DM cohort (data not shown); 7.4% of the patients in the high all-cause cost cohort had liver disease vs 1.4% of the patients in the lower-cost all-cause cohort (Table 2). Similarly, insulin use during the study period, a driver of both all-cause and T2DM-specific costs, has been more frequent in the high cost than lower-cost all-cause and T2DM costs cohorts (40.9% vs 18.3% for all-cause cost and 52.9% vs 14.3% for T2DM costs). This suggests that hypoglycemia, liver disease and insulin use are associated with utilization of expensive health services and explains their strong impact on the patients’ odds of being in the respective high cost cohorts.

The results of this study may not fully apply to the general population of T2DM patients. First, the findings of this study apply mainly to T2DM patients under 65 years of age. Second, patients covered by capitated or partially-capitated insurance plans were excluded because they received services whose costs are not captured by the fee-for-service claims; however, this is not expected to affect the generalizability of our results (see Supplementary eDiscussion1 for a comparison with patients covered by HMO/POS plans). Third, this study was restricted to patients with two or more HbA1c lab measurements in the 6 months before and after the index date to allow the assessment of glycemic control as a potential cost driver. However, our conclusions did not change when we relaxed this criterion to include patients with one lab test over the same period (see Supplementary eDiscussion1 for specific results). Fourth, some differences in the geographical distribution of patients in the Lab sub-sample are expected, depending on the health plans that report lab information, but this is addressed by controlling in the analyses for the region of residence and patient comorbidities. Finally, the results of the current analysis do not apply to patients only receiving insulin therapy, as one of the selection criteria was that patients must be taking some form of OHAs.

Other study limitations include omissions and coding errors, which are possible in healthcare claims data. However, these are not expected to be systematic between the study cohorts. There were also several limitations specific to this study. First, the study reports a conservative estimate for the relative share of diabetes-related costs relative to total costs, as diabetes-related costs in this study required claims associated with a T2DM diagnosis or a hypoglycemic agent. Second, in the current study the duration of the disease was not assessed. Because the cost of T2DM due to medications, complications, and admissions may increase as the duration of T2DM increases28–30, the costs measured in this study would be under-estimated if the proportion of patients with short duration of disease from this study sample is larger than the proportion of patients with short duration of disease in the general population of T2DM patients. However, the duration of disease in the study sample is likely to be representative to that of the general T2DM population, as the index date used for each patient was randomly selected from all dates with an OHA prescription fill. Third, the results of our analyses depend on the cut-offs chosen to define high cost patients; however, we believe using the third quartile as a cut-off provides a good trade-off between maximizing the models’ statistical efficiency and preserving a good discrimination between high and lower cost patients. Fourth, older age appeared to be associated with lower costs in the diabetes-related cost model, but this is due to the fact that the 3% of the patients aged ≥65 years have many of their costs covered by Medicare. Fifth, our ability to assess the impact of rare specific drivers was limited. Sixth, the causal effect of highly correlated cost contributors is difficult to separate, leading to possible under-estimates of the effect of certain factors. Finally, patients’ average HbA1c levels during the study period were imputed using the last value carried forward, given that it is not fully understood how patient HbA1c levels change between tests.

Conclusions

This study found that complications of diabetes and comorbidities were the most important factors leading to high healthcare costs, with IP admissions for complications of diabetes being the strongest predictors of both all-cause and diabetes-related costs. Overall, the findings of our analysis provide fresh insights into potential targets for the development of new hypoglycemic agent treatments and may inform decisions aimed at reducing the escalating costs of diabetes.

Transparency

Declaration of funding

This study was sponsored by Takeda. The study PI was AG, from Analysis Group.

Declaration of financial/other relationships

This study was sponsored by Takeda. MB, CL, and PV are employees of Takeda Pharmaceuticals International, Inc. and own restricted stock options. AG, DLV, RII, and EW are employees of Analysis Group, Inc., which has received consultancy fees from Takeda Pharmaceuticals International, Inc. Analysis Group is an economic consulting firm that is engaged by a variety of clients in the bio-medical arena for a range of services including the development of scientific communication materials for health economics and outcomes research across many therapeutic areas. Analysis Group’s services for each client are confidential and independent of services for other clients.

Acknowledgments

The authors thank Ana Bozas, PhD, who is an employee of Analysis Group Inc., for her contributions to the preparation and editing of this manuscript.