Evaluation of cardiovascular risk factors, events, and costs across four BMI categories

Authors


  • Disclosure: The authors declared no conflict of interest.

  • Funding agencies: This research was supported by grant from Takeda Pharmaceuticals International, Inc.

  • Present address of J. Yu: Touro University California College of Pharmacy, 1310 Club Drive, Mare Island, Vallejo, CA 94592.

  • Present address of S. Raparla: HealthCore, Inc., 800 Delaware Avenue, Fifth Floor, Wilmington, DE 19801.

Abstract

Objective

The purpose of this study was to analyze the 1-year cost of cardiovascular (CV) events by body mass index (BMI) subgroups from a US employer health plan perspective.

Design and Methods

Patients aged 20-64 years from the GE Centricity Electronic Medical Record, National Health and Nutrition Examination Survey, and MarketScan databases were used to determine prevalence of risk factors (RFs) and CV events and 1-year costs. Risk factors included hypertension (HTN), diabetes, and hyperlipidemia (HLD) and CV events included myocardial infarction, angina, heart failure, and stroke. CV event costs were determined from claims by ICD-9 code in patients with overweight/obesity.

Results

Of 220,136 patients identified in GE, BMI was 25-26.9 in 19.4%, 27-29.9 in 30.4%, 30-34.9 in 27.9%, and ≥35 in 22.3%. Patients with diabetes, HTN, and HLD increased with BMI from 1.8% (25-26.9) to 11.4% (≥35) in males and 1.1% to 6.8% in females. Prevalence of CV events increased from 0.1% with no RFs up to 10.2% with multiple RFs. The average 1-year cost per patient increased from $1122 to $2383 as BMI increased.

Conclusions

Patients with higher BMI values had an increased prevalence of RFs and CV events, which lead to higher average 1-year costs.

Introduction

Cardiovascular (CV) disease is one of the leading causes of mortality and significantly impacts the cost of healthcare in the United States [1]. The American Heart Association estimated the total indirect and direct costs of CV disease in the United States to be $503 billion in 2010 [1]. The National Cholesterol Education Program-Adult Treatment Panel (NCEP-ATP III) guidelines recognize many CV disease risk factors (RFs) including hyperlipidemia (HLD), hypertension (HTN), diabetes mellitus (DM), and obesity [2]. Additionally, these CV RFs have also been associated with increased CV disease mortality [3, 4]. Overweight and obesity play a major role in adversely affecting CV disease RFs and have also been considered probable independent RFs for CV events [5]. Body mass index (BMI) in kg/m2 is an indicator that has been widely used as a measure for overweight and obesity, and increasing BMI has been related to CV health risks [5, 6]. According to the National Heart, Blood and Lung Institute guidelines, a BMI of 25-29.9 is considered overweight, BMI of 30-34.9 is class I obesity, and BMI ≥35 is a combination of class II and III obesity [7]. Obesity (BMI ≥30) is increasing among adults in the United States and in 2007-2008, the overall age-adjusted prevalence of obesity was 34%, with 32% among men and 36% among women [8]. Furthermore, 68% of adults in the United States are overweight or obese (BMI ≥25) [8] and overweight individuals (BMI 25-29.9) make up a large part of the workforce at 38% with obese individuals comprising 29% [9].

Overweight and obesity imposes substantial costs to employer health plans in the United States. Medical spending attributed to overweight and obesity accounted for 8.2% of total spending, between $19 and $28 billion, for private insurance in 1998 [10]. Furthermore, overweight and obese full-time employees have higher medical expenditures and rates of absenteeism than normal weight individuals [9, 11] and extreme obesity (BMI ≥35) has been associated with decreased productivity [12]. Many articles have been published examining the impact of obesity on the development of CV RFs and outcomes [13], the impact of RFs on CV events [17], and costs of RFs [18, 22], but currently there is no literature examining the impact of BMI on the prevalence of specific combinations of CV RFs, CV events, and, finally, costs to employer health plans using observational data. Such data is important in understanding the economic impact of overweight and obesity in CV disease, which may in turn be used to help make coverage decisions. The purpose of this study was to analyze the cost of obesity in terms of CV events using observational data from the perspective of employer health plans.

Methods and Procedures

A decision tree was used for the analysis of three nationally representative sources which were used for primary data inputs. These included the General Electric Centricity Electronic Medical Record (GE EMR), the National Health and Nutrition Examination Survey (NHANES), and the Thomson Reuters MarketScan databases. The GE EMR was used to link BMI groups to the CV RFs HTN, HLD, and DM. Due to the ambulatory care nature of the GE EMR, there may be underreporting of CV events and, thus, NHANES was used to link CV RFs to CV events. However, NHANES does not have cost data and, therefore, MarketScan, an insurance claims database, was used to link CV events to medical costs.

GE EMR: Demographic and risk factor probabilities

The GE EMR (Waukesha, WI) was used to determine the probability inputs for demographic (sex) and CV RFs (total number and combination). The GE EMR research database contains de-identified, longitudinal ambulatory care electronic health data contributed by over 15,000 providers located in over 35 states for 15 million patients. Data include but are not limited to demographic information, vital signs, laboratory orders and results, medication list entries and prescription orders, and diagnoses or problems. The use of these data has enabled research of treatment outcomes based on clinical parameters [13, 23].

Patients were stratified by BMI according to the NHLBI guidelines [7]. However, because treatment of overweight individuals should be considered at a BMI of ≥27 when other RFs are present [24], the GE EMR population was stratified into four BMI groups: 25-26.9, 27-29.9, 30-34.9, and ≥35. Included patients were aged 20-64 years on the index date and had a recorded BMI ≥25 between March 1, 2005 and June 30, 2009. The index date was defined as the date of the recorded BMI (or latest BMI if multiple values were present). To reduce the risk of classification bias, all BMI values were required to be in the same BMI group. Patients were required to have a minimum of 395 days of EMR activity before the index date to maximize the capture of medication use. The NCEP-ATP III and Joint National Committee on the Prevention, Detection, Evaluation, and Treatment of High Blood Pressure (JNC 7) guidelines recognize males have a higher risk of CV disease at a younger age, so each BMI group sample was stratified by sex [2, 25]. Each BMI-sex group was stratified by number of CV RF (0, 1, 2, or 3 RF). The RF included DM (both type 1 and type 2 DM), HTN, and HLD and were identified by ICD-9 diagnosis codes (within 2 years prior to the index date), prescription medication orders for antihypertensives, antihyperlipidemics, or antidiabetics by generic product identifier (within 395 days before the index date), or clinical lab values (within ±90 days of the index date). For clinical lab values, patients were considered to have the RF if they were not at goal for DM—HbA1C ≥7% [26], HTN—systolic blood pressure (SBP) ≥140 mmHg or diastolic blood pressure (DBP) ≥90 mmHg for nondiabetic patients or SBP ≥130 mmHg or DBP ≥80 mmHg for diabetic patients [25], or HLD—total cholesterol (TC) ≥240 mg/dL, high density lipoprotein cholesterol (HDL) <40 mg/dL, low density lipoprotein cholesterol (LDL) ≥160mg/dL, or triglycerides (TG) ≥200 mg/dL [2]. Each RF group was further stratified by the individual RF combinations (0 RF; 1 RF: only DM, HTN, or HLD; 2 RF: DM + HTN, DM + HLD, HTN + HLD; all 3 RF).

NHANES: CV event probabilities

The 2007-2008 NHANES data was used to determine the prevalence of nonfatal CV events in patients with the specific combinations of DM, HTN, and HLD. The NHANES is a program of studies done by the Centers for Disease Control and Prevention. The NHANES was initiated in the early 1960s and brings together both interview and physical examination data. The survey captures the health and nutritional data of about 10,000 adults and children each year, who are representative of the United States population. The interview data from NHANES contains demographic, socioeconomic, and dietary information along with the health-related data [27].

The NHANES 2007-2008 database was used to estimate the probability of CV events in the decision tree. Five possible CV event outcomes were included: no event, myocardial infarction (MI), angina, heart failure (HF), and stroke. The prevalence of CV event outcomes was determined by questionnaire data from patients 20-64 years of age with BMI ≥25 for each of the RF combinations determined from the GE EMR. Initially, CV event prevalence was to be determined for each BMI-sex-number of RF-RF combination, but the BMI and sex stratifications were collapsed due to missing data. Prevalence of RF and CV events were determined by the question, “Has a doctor ever told you that you had [x]?” where x was MI, stroke, angina, HF, DM, HTN, or HLD. To provide the most conservative estimate of RF and CV event prevalence, patients that did not know, refused to respond, or the response was unknown were classified as not having the specific CV event or RF. It was assumed that each event outcome was from a unique patient and the total patient population was equal to the sum of patients with event responses. For example, the probabilities of having an MI used in the decision tree were determined by dividing the number of patients with an MI by the total number of patients with event responses (MI + angina + HF + stroke + no event). Thus, the denominator was greater than the actual number of patients responding to the survey and underestimated the prevalence of each CV event. This approach did not account for patients with multiple events. Furthermore, data from patients ≥20 years with BMI ≥25 was used for the CV event rates in the RF combinations of DM and DM + HLD due to missing data in the 20-64 age groups.

MarketScan: CV event and RF costs

The MarketScan database was used to determine the 1-year cost of RF and CV events. The Thomson Reuters MarketScan Research Databases contain individual-level, de-identified, healthcare claims information from employers, health plans, hospitals, and employer sponsored Medicare supplement programs. Among the databases available for study are the Commercial Claims and Encounters database and the Medicare Supplemental and Coordination of Benefits database [28, 29].

The Commercial Claims and Encounters database contains data for over 34 million patients and was used to determine the 1-year costs of the CV events reported from the NHANES and RF from the GE EMR. Patients were included if they had an ICD-9 code for overweight, obesity (unspecified), or morbid obesity and were 20-64 years of age between July 1, 2005 and June 30, 2006. CV events and RF were identified by the presence of an ICD-9 code (defined as the index date) between the last overweight or obesity claim and June 30, 2008. Costs were extracted from inpatient and outpatient claims when the CV event or RF was the primary or secondary ICD-9 code, the patient was enrolled for at least 365 days after the index date, and, for CV events, the patient had at least one inpatient claim. To focus on event-based costs, prescription drug costs were not included. Costs were then determined for the 365 days following the index date, and therefore, only nonfatal CV events were considered. To minimize bias and cost outliers, the lower and upper 0.01% of cost data was removed from the analysis.

Risk algorithm

A decision tree was developed to present the relationships between RFs, CV events, and costs across BMI groups. This was done using Microsoft Office Excel® 2007 and TreeAge Pro© 2009. The first branches from the decision node were the four BMI groups (25-26.9, 27-29.9, 30-34.9, and ≥35). The first set of chance nodes in the decision tree was the stratification of each BMI group by sex. The probabilities for these branches were determined by the percentage of males and females in each BMI group from the GE EMR. The second set of chance nodes in the decision tree was the stratification of each BMI-sex combination by the number of RF (0, 1, 2, or all 3 RF). The probabilities for these branches were determined by the percentages of the number of RF present within each BMI-sex combination from the GE EMR. The next set of chance nodes was the BMI-sex-number of RF groups stratified by the specific combination of RF present (0 RF; 1 RF: DM, HTN, or HLD; 2 RF: DM + HTN, DM + HLD, or HLD + HTN; 3 RF: DM + HLD + HTN) and percentages from the GE EMR were again used as probability inputs. The final set of chance nodes was the probabilities of the five CV event outcomes for each RF combination. The probability inputs for these branches were determined from prevalence data from the NHANES database. Finally, the 1-year cost of a CV event, determined from the MarketScan database, was used as the payoff (cost) values for the decision tree. Because the ICD-9 codes in MarketScan did not specifically match the BMI groups considered from the GE EMR, CV event costs were calculated from patients with a diagnosis of either overweight or obesity. These costs were assumed to be equal across each BMI category to provide a conservative estimate.

The expected value of each branch of the tree was determined by multiplying the probability of that branch by the expected cost of that branch. The value of each node was determined by adding the expected values of each of the branches stemming from that node. An illustration of the decision tree is shown in Figure 1.

Figure 1.

Structure of base case scenario decision tree. The model structure shows the base case analysis decision tree, which did not consider the 1-year cost of risk factors and thus assumed the 1-year cost of no event was $0. The numbers in the boxes indicate the smaller tree with the corresponding number stems from that node. Uncertainty in the cost parameters was addressed in the second and third scenarios (model structures not shown). Abbreviations: BMI, body mass index; HTN, hypertension; DM, diabetes mellitus; HLD, hyperlipidemia; CV, cardiovascular; cNoEvent, 1-year cost of no CV event; MI, myocardial infarction; cMI, 1-year cost of MI; cAngina, 1-year cost of angina; HF, heart failure; cHF, 1-year cost of HF; cStroke, 1-year cost of stroke.

Sensitivity analyses

A one-way sensitivity analysis varies only one parameter in the decision tree over a range of values at a time. One-way sensitivity analyses were conducted on both the number of RF and CV event probabilities used in the decision tree. The 95% confidence intervals (95% CI) were used for the upper and lower values in the one-way sensitivity analyses. During each one-way sensitivity analysis, it was assumed when an increase or decrease in the number of RF or CV event probabilities occurred the difference would be shifted to or from the no RF or no event groups. For example, when the probability of having 1 RF in the BMI 25-26.9 male subgroup was varied over the 95% CI, the probability of having 0 RF adjusted accordingly to ensure the sum of the probabilities of the number of RF branches was equal to 1.

The 1-year cost of RF and CV events did not discriminate by specific combinations of RF so three scenarios were designed to address uncertainty of cost in the decision tree. In the base case, the cost of the RF was not addressed, and thus, the cost of no event was $0 and the cost of a CV event was simply the cost of the event. However, it is unrealistic to assume a patient with RF would not incur any cost in the absence of a CV event; therefore, two additional analyses were performed. In the second analysis, it was assumed that the cost of the CV event already included the cost of the RF and the cost of no event was the additive cost of RF in the pathway (e.g., in a patient with DM + HTN and no CV event, the cost of no CV event was equal to the 1-year cost of DM and 1-year cost of HTN). For the third analysis, it was assumed the cost of RF was not included in the cost of the event, and thus, the 1-year cost was the cost of the event plus the additive cost of any RF and the cost of no event remained the additive cost of RF (e.g., in a patient with DM + HTN and an MI, the cost of the MI was the 1-year cost of DM, HTN, and MI).

This study was reviewed and considered exempt by the IRB at the University of Utah.

Results

The patient flowchart from the GE EMR study population is shown in Figure 2. There were over 4.6 million patients with at least one activity from March 1, 2005 to June 30, 2009 identified from the GE EMR. Of these, over 1.5 million were ≥20 years of age and had only one BMI or ≥2 BMI values in the same group. Just over 1 million patients had a BMI ≥25 and of those 847,979 were 20-64 years of age. Finally, the analysis was performed on the 220,136 patients remaining who met the inclusion criteria of having ≥395 days of pre-index date EMR activity.

Figure 2.

Patient flow chart from GE EMR. The flow chart shows the steps in identifying the patient population from the GE EMR database and the number of patients in each step. Abbreviations: GE, General Electric Electronic Medical Record Database; BMI, body mass index.

The probability inputs determined from the GE EMR database for males and females in each BMI group, excluding the specific RF combinations, are presented in Table 1. There were 19.4% with BMI 25-26.9, 30.4% with BMI 27-29.9, 27.9% with BMI 30-34.9, and 22.3% with BMI ≥35, which was similar to the overall GE EMR population with recorded BMI values (data not shown). Overall, 55.2% of the patients were female. In the overall male population, 44.7% had 0 RF, 34.8% had 1 RF, 15.9% had 2 RF, and 4.7% had all 3 RF. In the overall female population, 62.2% had 0 RF, 25.2% had 1 RF, 9.2% had 2 RF, and 3.3% had all 3 RF. In females, the percentage of patients with 1, 2, or 3 RF increased and the percentage of patients with 0 RFs decreased as BMI increased. In males, the percentage of patients with 2 or 3 RF increased and the percentage of patients with 0 RF decreased as BMI increased. In males, however, the percentage of patients with 1 RF decreased in the ≥35 BMI group.

Table 1. GE EMR—risk factor probabilities by sex and BMI group
VariableBMI 25-26.9BMI 27-29.9BMI 30-34.9BMI ≥ 35
(N = 42,680)(N = 66,908)(N = 61,451)(N = 49,097)
N%95% CIN%95% CIN%95% CIN%95% CI
  1. Abbreviations: 95% CI, 95% confidence interval; BMI, body mass index; DM, diabetes mellitus; GE EMR, General Electric Electronic Medical Record Database; HLD, hyperlipidemia; HTN, hypertension.
Male19,86946.6% 33,02349.4% 29,23947.6% 16,54233.7% 
Female22,81153.4% 33,88550.6% 32,21252.4% 32,55566.3% 
Overall prevalence
DM12923.0%[2.9-3.2]25703.8%[3.7-4.0]43207.0%[6.8-7.2]682613.9%[13.6-14.2]
HTN946422.2%[21.8-22.6]18,63027.8%[27.5-28.2]22,93837.3%[36.9-37.7]23,31447.5%[47.0-47.9]
HLD849719.9%[19.5-20.3]15,85523.7%[23.4-24.0]17,25128.1%[27.7-28.4]13,66227.8%[27.4-28.2]
Male
0 Risk factors11,09755.9%[55.2-56.5]16,26349.2%[48.7-49.8]11,55239.5%[38.9-40.1]515431.2%[30.5-31.9]
1 Risk factor635532.0%[31.3-32.6]11,42434.6%[34.1-35.1]10,71036.6%[36.1-37.2]580435.1%[34.4-35.8]
2 Risk factors205510.3%[9.9-10.8]446613.5%[13.2-13.9]5,48018.7%[18.3-19.2]370622.4%[21.8-23.0]
3 Risk factors3621.8%[1.6-2.0]8702.6%[2.5-2.8]1,4975.1%[4.9-5.4]187811.4%[10.9-11.8]
Female
0 Risk factors16,87574.0%[73.4-74.5]23,25268.6%[68.1-69.1]19,20959.6%[59.1-60.2]16,26950.0%[49.4-50.5]
1 Risk factor442819.4%[18.9-19.9]772022.8%[22.3-23.2]867226.9%[26.4-27.4]984630.2%[29.7-30.7]
2 Risk factors12505.5%[5.2-5.8]23707.0%[6.7-7.3]331710.3%[10.0-10.6]421412.9%[12.6-13.3]
3 Risk factors2581.1%[1.0-1.3]5431.6%[1.5-1.7]10143.1%[3.0-3.3]22266.8%[6.6-7.1]

The probability inputs determined from the NHANES database are presented in Table 2. Under the assumption that each event occurred in a unique patient, the sum total of patients with CV event outcomes used to determine the decision tree probabilities was 119.8 million. Overall, the percentage of each CV event increased as the number of RFs increased. The DM + HTN group had the highest event rate in HF (9.3%) and stroke (10.2%), whereas the DM + HTN + HLD group had the highest event rate in MI (9.8%) and angina (9.0%). When comparing CV event rates of RF combinations, patients with DM generally had higher rates of events than those without DM. However, because the event rate for the DM and DM + HLD groups was determined using patients ≥20 and not 20-64, due to sample size limitations this effect may be overestimated.

Table 2. NHANES—cardiovascular events probabilities by risk factor combination
Risk factorsTotal patientsaMIAnginaStroke
NN%95% CIN%95% CIN%95% CI
No risk factors62,318,040212,4900.34%[0.34-0.34]87,7410.14%[0.14-0.14]336,7210.54%[0.54-0.54]
DM onlyb2,587,512134,9565.22%[5.19-5.24]80,8813.13%[3.10-3.15]103,5464.00%[3.98-4.03]
HLD only16,054,103195,2271.22%[1.21-1.22]230,8641.44%[1.43-1.44]134,1480.84%[0.83-0.84]
HTN only16,012,420459,6252.87%[2.86-2.88]379,0712.37%[2.36-2.37]476,4202.98%[2.97-2.98]
DM + HLDc2,159,934126,3215.85%[5.82-5.88]107,0544.96%[4.93-4.99]44,3522.05%[2.03-2.07]
DM + HTN2,306,452204,8478.88%[8.84-8.92]156,6636.79%[6.76-6.83]235,98110.23%[10.19-10.27]
HTN + HLD13,633,111573,5434.21%[4.20-4.22]419,0073.07%[3.06-3.08]457,1153.35%[3.34-3.36]
DM + HLD + HTN4,722,822463,7149.82%[9.79-9.85]424,0668.98%[8.95-9.01]319,5696.77%[6.74-6.79]
Risk factorsTotal patientsaHFNo CV events
NN%95% CIN%95% CI
  1. Abbreviations: NHANES, National Health and Nutrition Examination Survey; MI, myocardial infarction; HF, heart failure; CV, cardiovascular;95% CI, 95% confidence interval; DM, diabetes mellitus; HLD, hyperlipidemia; HTN, hypertension.
  2. aAssumed each CV event was in a unique patient and thus is equal to sum of patients with MI, angina, stroke, HF, and no CV event.
  3. bThe event rates in these groups are for ≥20 years with BMI ≥25 population due to missing data in the 20-64 years group.
No risk factors62,318,04068,7820.11%[0.11-0.11]61,612,30698.87%[98.86-98.87]
DM onlyb2,587,512156,0556.03%[6.00-6.06]2,112,07481.63%[81.58-81.67]
HLD only16,054,103152,3330.95%[0.94-0.95]15,341,53195.56%[95.55-95.57]
HTN only16,012,420355,6542.22%[2.21-2.23]14,341,65089.57%[89.55-89.58]
DM + HLDb2,159,93429,3851.36%[1.34-1.38]1,852,82285.78%[85.73-85.83]
DM + HTN2,306,452214,8799.32%[9.28-9.36]1,494,08264.78%[64.71-64.84]
HTN + HLD13,633,111175,6641.29%[1.28-1.29]12,007,78288.08%[88.06-88.10]
DM + HLD + HTN4,722,822339,2347.18%[7.16-7.21]3,176,23967.25%[67.21-67.30]

A total of 246,261 patients with overweight or obesity claims were identified from the MarketScan database with 221,648 aged 20-64. There were 1,289 patients with HF, 1,204 patients with MI, 635 with stroke, and 539 with angina meeting the inclusion criteria from which costs were extracted. From the study population, the mean 1-year cost per patient of CV events and RFs used in the decision tree are reported in Table 3.

Table 3. MarketScan—mean 1-year cost of risk factors and cardiovascular events
CV event or risk factorMean 1-year cost (SD)
  1. Abbreviation: CV, cardiovascular.
Myocardial infarction$42,563 ($87,514)
Angina$5002 ($15,021)
heart failure$40,517 ($140,140)
Stroke$17,054 ($40,584)
Hypertension$1511 ($9347)
Diabetes$3896 ($13,925)
Hyperlipidemia$1136 ($5626)

From the decision tree analysis, the overall average 1-year cost per patient in each BMI group from the base case analysis showed the cost per patient increased as BMI increased (25-26.9—$1122; 27-29.9—$1323; 30-34.9—$1749; and ≥35—$2383). Tornado diagrams showing the spread of the most influential number of RF and CV events for each BMI group are presented in Figure 3. The one-way sensitivity analyses of the number of RF and CV events produced the same results as the base case in which the average cost per patient increased as BMI increased. The analyses revealed the number of RF with the most influence was all three RFs in males in all BMI groups (spread ranging from $15.60 to $25.50) except the BMI ≥35 group where females with all three RF (spread $27.80) was the most influential. The CV events had an even smaller impact on the average cost per patient results than RF. The most influential CV event in each BMI group was MI in patients with 0 RF (spread ranging from $1.90 to $2.50) except in the BMI ≥35 group where MI in patients with HTN was the most influential (spread $2.10). The three scenarios tested for the cost analysis also showed that as BMI increased the average cost per patient increased. These results are shown in Figure 4. As expected, the average cost per patient increased from the base case as the cost of no event changed from $0 to the cost of RF in the second scenario (ranging from $1702 to $3617). Also as expected, the third scenario in which the cost of comorbidities was added to the cost of no event and CV events produced the highest average per patient cost of the analysis (ranging from $1801 to $3958).

Figure 3.

Tornado diagrams for each BMI group. The tornado diagrams show the impact of variable uncertainty (ranged over 95% CI) for the five most influential number of risk factor and CV event variables (10 total) on 1-year costs in the model for each BMI group. The baseline expected value is represented by a dotted line. Abbreviations: BMI, body mass index; MI, myocardial infarction; HF, heart failure; HTN, hypertension; HLD, hyperlipidemia.

Figure 4.

Average 1-year cost per patient in each BMI group from cost analyses. The base case analysis did not consider the cost of risk factors and thus assumed the cost of no event was $0. Uncertainty surrounding the cost parameters was addressed in the second scenario where the 1-year cost of no event was considered to be the 1-year cost of the risk factors in the decision pathway and assumed the 1-year cost of CV events captured the cost of the risk factors. The third scenario added the 1-year cost of the risk factors to both the cost of no event and the cost of a CV event. Abbreviation: BMI, body mass index.

Discussion

The purpose of this study was to determine the cost of overweight and obesity with regard to CV events. To accomplish this, three different data sources were used to acquire the most appropriate inputs needed to conduct such an analysis at a national level. The decision tree cost analysis showed that patients with higher BMI values have higher costs, in terms of CV events, than patients with lower BMI values from the perspective of employer health plans using observational data. One of the strengths of this study is that it is the first to report costs of BMI groups with regard to CV events when considering specific combinations of RFs. The decision tree model could be used by employer health plans to see the 1-year costs of specific subgroups within the model (e.g., average cost of males with a BMI ≥35 and all 3 RF) to allow more targeted interventions for high-cost groups.

In the current analysis, the number of RF increased as BMI increased and the percentage of CV events increased as the number of RF increased. This led to higher costs in the higher BMI groups due to an increased number of CV events associated with an increased prevalence of RFs. The RF distribution in this study is supported by previous research in the GE EMR that also demonstrated an increase in the number of RFs as BMI increases [13]. The cost results from this study are comparable with the findings of other studies that examined the overall cost of obesity. In a study performed using the Medical Expenditure Panel Surveys (MEPS), obese patients with private insurance had medical expenditures $1140 greater than normal weight individuals in 2006 [30]. In the study, the authors analyzed MEPS data from 1998 and 2006 across Medicaid, Medicare, and private insurers using regression techniques and compared the results to patients with normal weight. In a separate study also using MEPS data, full-time employees across overweight and obese BMI groups had medical expenditures between $169 and $1591 higher than normal weight individuals [11]. The second study also used regression techniques to estimate the annual expenditures of men and women using MEPS data from 2000 and 2001. However, neither study looked specifically at the cost of CV outcomes, but at total spending on obesity.

The overall goal of the current study may have been approached several ways. One would be to do a prospective observational study; however, due to the significant amount of time between weight increase, development of RFs, and subsequent CV events, this would take many years to complete. Thus, a retrospective approach allows quicker access to information; however, there are limitations with this approach. There is no single retrospective database to capture all variables required. The most commonly used are administrative claims databases, which capture reimbursable medical and medication events through ICD-9 (diagnosis), current procedural terminology (CPT), and NDC (medication) codes. Such a database may collect some lab results but will not have any access to biometric data such as blood pressure or BMI. An EMR system allows for the capture of biometric data, along with diagnosis, medication orders and procedures; however, the data source is encounter driven, not claims driven. The current analysis included encounters from all providers, but the majority of providers (∼60%) in the GE EMR database are primary care providers. Thus, the EMR database is largely developed from ambulatory care physician practice sites and does not necessarily have information on encounters that took place in the hospital or other physician offices, such as specialists, unless these are reported back to the primary care practice by the physician or patient and documented in the record. This is demonstrated by the fact that the rate of CV events in the GE EMR is underreported compared with national estimates. Patients with HTN in the GE EMR report the 3-year incidence of MI at 0.6% and stroke at 0.2% (McAdam-Marx et al., unpublished work) compared with a 1-year incidence of ∼0.3%, for each, from national data [31]. Additionally, EMR research databases do not include cost data. An alternative approach is to go to the literature to obtain RFs, event rates, and costs with increasing BMI; however, the information published is generally not presented by BMI, age, sex, and RF categories. Although NHANES reports both BMI and prevalence of RFs, there was not enough data for each specific BMI group and RF combination to get all of the required inputs for the model. Thus, the strengths of each database were used to develop a decision tree based on the RF distribution across BMI categories using the GE EMR database. To obtain consistent data on CV events, which may be documented in the hospital setting, the NHANES database was used to estimate the probability of having a CV event based on the prevalence by RF combinations. To establish costs the MarketScan administrative claims, database was used to look at 1-year costs associated with the CV events of interest. These inputs were laid out in a decision tree to establish the events and costs associated with different BMI and sex groups.

The decision tree model used in the current study only considered overweight or obese patients and a normal weight group was not used as a reference. A normal weight BMI group may have been an ideal comparator, but was excluded due to the difficulty of accurately capturing costs for this group. Whereas 68% of the United States population is overweight or obese [8], only 0.7% of the MarketScan population had an ICD-9 code for overweight or obesity. Thus, it cannot be reasonably assumed that the costs of CV RFs or events from patients without an ICD-9 code for overweight or obesity actually represents the normal weight population. Therefore, the group closest to normal weight, which represents a large portion of patients with private insurance, was used as a comparison group and the decision tree demonstrated that costs for patients with higher BMI values increased as BMI increased. However, previous work found the mean overall annual cost was $413 per patient for those <65 years with a BMI 18.5-27 and at least one cardiometabolic RF (Brixner et al., unpublished work). This was compared with $1,433 per patient in those with a BMI ≥27 which is similar to what was found in the current study (Brixner et al., unpublished work). These results provide a relative comparator group for the normal BMI group in the current study and support the conclusion that costs increase as BMI increases.

The decision tree did not include patients ≥65 years. However, the perspective chosen for the model was that of an employer health plan and since these patients would likely be covered by Medicare, they were not included in the model. Age is a significant RF for CV RFs and events and patients ≥60 years have higher rates of overweight and obesity than younger patients [8], but weight loss may also occur as part of the aging process [32]. Had patients ≥65 been included in the model, the average cost per patient in each BMI group would have likely increased due to increased risk (i.e., higher rates of RFs, CV events, and death). Although a similar distribution of patients among the BMI groups may have been seen [8], the impact of weight on CV RFs, events, and costs remains unknown. The elderly population warrants separate examination in future studies because they are inherently different than the younger patients included in the model.

Limitations

The decision tree used in the current study has several limitations that arose from the need to use multiple data sources to acquire all of the necessary inputs. First, the CV event probabilities from NHANES used in the decision tree only considered prevalence, not incidence and do not take into consideration the timing of events. This may cause an overestimation of the 1-year costs because of an increased number of events. However, patients in the GE EMR were identified based upon encounters with providers which may underestimate the actual number of patients in each BMI or with RFs. However, NHANES is a nationally representative database and was used to determine the number of patients with each specific combination of RFs with a CV event.

The methods for determining the probability of a CV event assumed each CV event was from a unique patient and did not account for patients with multiple events. Thus, it remains unclear how multiple events would impact the prevalence of CV events, and the decision tree is unable to address the impact of multiple events on 1-year costs. Because NHANES is self-reported data from a questionnaire, the responses may be subject to recall bias. However, NHANES has been used extensively by researchers to determine prevalence rates of diseases in the United States. As NHANES is a nationally representative sample and the GE EMR represents patients seeking ambulatory care, including self-pay individuals, the prevalence rates of CV RFs and events may be different than an employer health plan population. However, the actual impact this may have on the model remains unknown because the distribution of age and race and the prevalence of HTN and DM is similar between the GE EMR and national estimates [33]. Finally, only patients in a single BMI group were included from the GE EMR, and thus, the decision tree is unable to assess the impact of changes in BMI and how this might affect costs of RFs and CV events.

Conclusions

This work demonstrated the potential relationship between obesity, CV RFs, CV events, and their costs, which can provide important information to employer health plans making decisions on the management of obesity and its related risks and costs. This study was not designed to examine the impact of changes in BMI on costs, but does suggest that patients in lower BMI groups cost less. Further studies are warranted to examine the hypothesis that treating patients with RFs and a BMI ≥27 may reduce the cost of CV events to employer health plans. Future research examining the impact of age on the cost of obesity and CV events is suggested. Future research in this area is warranted to determine if change in BMI over a lifetime may have a long-term impact on the cost of CV events.

Acknowledgments

Takeda Pharmaceuticals International, Inc. provided funding for this study. D.I. Brixner has received financial compensation for consulting from Bristol-Myers Squibb, Novo Nordisk, and Teva Pharmaceuticals. M. Bron is an employee of Takeda Pharmaceuticals. G.M. Oderda has received financial compensation from Novo Nordisk for board membership, manuscript preparation, and travel to meetings and received fees for speaking from Janssen.

Ancillary