Change in Medical Spending Attributable to Diabetes: National Data From 1987 to 2011

In the past 2 decades, the number of Americans in whom diabetes has been diagnosed more than doubled (1). Along with the growth in the prevalence of diabetes, the complexity of diabetes care has also substantially increased (2). This is, in part, due to the translation into clinical practice of findings in landmark trials such as the UK Prospective Diabetes Study (3) and the Diabetes Control and Complications Trial (4). Those studies, conducted between the late 1980s and early 1990s, established the microvascular benefits of tight glycemic control and the macrovascular benefits of tight blood pressure and lipid control for adults with diabetes (4,5). After the publication of those studies, increasingly aggressive approaches were gradually adopted in diabetic management (6,7). Considerable development and diffusion of new pharmacological products and medical procedures followed. Patients now receive a more complicated management regimen to reduce the risk of macrovascular and microvascular complications.

Not surprisingly, these changes have led to significant ramifications for the pattern of medical use and spending and thus warrant an examination of the change in diabetes-related costs over time. Previous studies have documented a considerable increase in the use of oral medications (8) and the number of prescription medications used among diabetes patients in managed care and ambulatory care settings (9,10). However, any changes in the use of and spending on hospitalizations, outpatient visits, and other medical services related to diabetes care remain unclear. Few studies have comprehensively quantified the change in the use of all these medical components in the U.S. Moreover, no study has used national-level data to examine the trend of the financial burden of diabetes from the patients’ or payers’ perspectives.

In this study, we assembled data on inpatient, outpatient, and emergency room (ER) visits, as well as prescription medication use, reported in two national data sets, to examine how medical spending and use attributable to diabetes have changed between 1987 and 2011. These nationally representative data provide a comprehensive assessment of the cost impact related to the changes in diabetes care.

We used data from the 1987 National Medical Expenditure Survey (NMES) and data from the 2000–2001 and 2010–2011 Medical Expenditure Panel Survey (MEPS). NMES, the predecessor of MEPS, was conducted among 22,538 individuals only in 1987. MEPS is a set of national surveys of families and individuals, their medical providers, and employers in the U.S. It samples data on an average of 39,000 individuals per year to estimate the use of medical resources in the U.S. population. We used the most recent 2-year data, from 2010 to 2011 along with prior 2-year data collected a decade earlier (2000–2001). Both the NMES and MEPS use the same method to collect information on health care use, expenditures, sources of payment, health status, the status of health insurance coverage, and demographic and socioeconomic characteristics of the civilian, noninstitutionalized population of the U.S. Because of the change in the charge-to-cost ratio over time, we adjusted medical expenditures reported in the NMES following the data providers’ guide (11) to make it comparable to the MEPS. Medical use and expenditures were collected from both household respondents and their medical providers.

We identified diabetes cases by respondents’ self-reported physician’s diagnosis. In the 2010–2011 MEPS, the question concerning diabetes diagnosis was asked only among persons ≥18 years of age. Therefore, we included diabetes patients ≥18 years of age in the study sample to ensure consistency over study periods. In both the NMES and MEPS, medical expenditures were reported as the total payment for health care services by all payers (including insurance, survey respondents, and other sources). Expenditures were measured in the following five categories: inpatient, outpatient, ER visits, prescription medications, and “other medical services.” The other medical services include medical equipment such as testing supplies, dental care, home health care, vision aids, and other miscellaneous items or services. Total medical spending was calculated as the sum of the spending on the five components. All spending was adjusted for inflation and expressed in 2012 dollars.

Medical use was measured in the following five categories: the numbers of outpatient visits, including office visits and hospital-based outpatient visits; ER visits; hospital admissions; nights of hospital stay; and prescription medication purchases (i.e., the total number of encounters with medication purchases, including initial purchases and refills). Medical spending per encounter with each medical service (e.g., spending per encounter of outpatient visit) was calculated by dividing medical spending by the quantities of the medical service used (e.g., the number of outpatient visits over a year).

We used a two-part model to estimate the medical spending attributable to diabetes. The model accounted for the large number of observations with no expenditures (i.e., zero costs) in the data. The two-part model included a logit model to estimate the probability of an individual having any positive medical spending, and the second part, for positive medical spending, was a generalized linear model with a log-link for annual medical spending. We performed a modified Park test (12) and determined that a γ-variance function was an appropriate variance specification for the generalized linear model. Medical uses were estimated using a zero-inflated negative binomial model. The model included a logit regression for predicting the probability of having zero use and a negative binomial count regression to fit the positive count data. We conducted a Vuong specification test and determined that a negative binomial distribution was a better fit than a Poisson distribution to model the count data (13). In all regression models, we included the following key explanatory variables: diabetes status, time period indicators, and their interaction terms. To account for the secular change in population demographics and socioeconomic position of people with or without diabetes, we also included the following covariates: age and age squared, and the interactions between diabetes diagnosis and age and between diabetes diagnosis and age squared, sex, race/ethnicity, BMI, residential regions, marital status, insurance coverage, educational attainment, and annual income. We also adjusted for other self-reported chronic medical conditions including high cholesterol, arthritis, current smoking, asthma, hypertension, and any current cancer. In this way, our analysis accounts for the secular change in the population-level patient characteristics; therefore, the estimated costs of diabetes are directly comparable across time periods. Because diabetes is a primary risk factor for cardiovascular diseases (CVDs), we did not include CVD to prevent potential underestimation of the effect of diabetes on medical spending (14,15). In a sensitivity analysis, we included CVD in the regression to obtain the lower-bound estimate of the spending attributable to diabetes.

The excess medical expenditures and uses attributed to diabetes were estimated by comparing the means of those outcomes that were predicted based on the above regression models by diabetes status. It represents the increased medical spending or use for diabetes patients, compared with what would have been spent or occurred if those patients with diabetes did not have diabetes. The SEs were estimated using 1,000 iterations of bootstrapping. In examining the spending growth, we presented both absolute and relative changes, and we emphasized absolute changes because, first, it provides a better measurement of the contribution of each medical component to the increase in total spending, and, second, compared with relative changes, absolute changes are less sensitive to the estimation errors of baseline spending, especially when the values of the spending are small. All statistical analyses were conducted using Stata, version 13 (StataCorp, College Station, TX).

Table 1 shows the characteristics of adults with and without diabetes in the U.S. during 1987, 2000–2001, and 2010–2011. Diabetes patients were, on average, older; less likely to be a non-Hispanic white or a smoker; more likely to have hypertension, CVD, and stroke; and more likely to be covered by health insurance than persons without diabetes. Despite those differences, similar trends were observed in people with and without diabetes between 1987 and 2011, as follows: the mean age for both groups slightly increased; the proportion of minorities for both groups increased; the prevalence of current smokers for both groups decreased; and the prevalence of hypertension for both groups increased. However, diverging trends exist. Adults without diabetes experienced a significant increase in the prevalence of CVD, whereas, among adults with diabetes, the prevalence of CVD remained unchanged. Additionally, adults with diabetes bore a greater increase in their (unadjusted) annual medical spending.

Table 2 shows that, after adjusting for observed differences, the estimated excess medical spending attributed to diabetes increased from $2,588 (95% CI $2,265 to $3,104) per person in 1987 to $4,205 (95% CI $3,746 to $4,920, P < 0.001) per person in 2000–2001, and to $5,378 (95% CI $5,129 to $5,688, P < 0.001) per person in 2010–2011. The increase was observed across all components of medical spending, with the greatest absolute increase in the spending on prescription medications ($1,528 increase), followed by inpatient visits ($680 increase) and outpatient visits ($430 increase). The absolute change in the spending on ER and other medical services use was relatively small. In relative terms, the spending on ER visits grew more than five times, faster than that of prescription medication and other medical components. Among the total absolute increase ($2,790), prescription medication accounts for 55%; inpatient visits account for 24%; outpatient visits account for 15%; and spending on ER visits and other medical services account for 6%. Among the total annual diabetes-attributable medical spending, the spending on inpatient and outpatient visits dropped from 40% and 23% to 31% and 19%, respectively, between 1987 and 2011, whereas spending on prescription medication increased from 27% to 41%. Accounting for CVDs, the estimated annual medical spending attributable to diabetes decreased by 14–17%, depending on the study period.

Table 3 shows that the trend of medical use diverged by the type of medical service. The use of prescription medication, measured by the number of total encounters of medication purchases, continuously grew by 51% between 1987 and 2011, whereas the use of other medical services either declined or remained relatively unchanged. The number of inpatient and ER visits declined both in adults with and without diabetes; yet, diabetes patients had a greater decrease. As a result, the frequency of inpatient and ER visits attributable to diabetes declined by 29% and 19%, respectively. The length of hospital stays attributable to diabetes decreased by half. Outpatient visits attributable to diabetes remained unchanged after a slight increase between 1987 and 2001.

The unit costs rose universally in all five measures of medical care in adults with and without diabetes. For each hospital admission, diabetes patients spent significantly more than persons without diabetes. The gap increased from $1,028 to $1,605 per hospital admission between 1987 and 2001, and dropped slightly to $1,360 per hospital admission in 2011. Diabetes patients also had higher spending per ER visit and per purchase of prescription medications. However, the spending per night of hospital stay and per outpatient visit was not significantly different by diabetes status. In diabetes patients, the greatest absolute growth of unit cost was observed in inpatient visits, followed by ER visits, prescription drugs, and outpatient services. In relative terms, ER visits experienced the most rapid growth (930%), followed by prescription medications (182%), hospital admissions (88%) and stays (172%), and outpatient visits (18%).

Diabetes care in the U.S. has changed substantially since the publication of landmark studies (2,16) 2 decades ago. Our study is one of the first studies to assess how those changes have affected patient-level medical uses and expenditures using nationally representative data. Our analyses reveal a doubling of the medical spending attributable to diabetes per person between 1987 and 2011. Over half of the spending increase was due to the growth in prescription medication spending, resulting from the increase in both use and price in prescription medications. The increase in outpatient expenditure was almost entirely driven by more outpatient visits. In contrast, the increase in inpatient and ER expenditures was caused by the rise in unit cost, as the number of hospital admissions, nights of hospital stay, and ER visits decreased modestly or remained unchanged.

Our results resemble the findings from previous studies with respect to the growing economic burden of diabetes in the U.S. (17–21), though discrepancies exist. A series of American Diabetes Association (ADA) studies from 1997 to 2012 suggested that the per capita medical cost attributable to diabetes changed from $11,759 (in 2012 dollars) in 1997 to $7,888 in 2012 (17–20). Because of the changes in the data used and the refinements in the analytical methods in those studies, it remains unclear whether the per capita medical spending was comparable over time. Because we applied the same analytical framework to national data sets that were collected using the same methods, our estimates account for many population-level changes over time and thus are directly comparable. Our estimated absolute values of the per capita medical spending also appear to be lower than the ADA estimates. This could be due to differences in estimation methods and data sources. The population etiological fractions method adopted by the ADA studies has been shown to produce somewhat higher estimates than those produced from the individual-based regression method used in the current study, probably due to the underestimation of the individual-level differences by diabetes status in the etiological fraction method (14). In addition, the ADA studies include national nursing home and home health care data, which are not included in the MEPS or NMES. The spending on those cases was shown to account for ∼10% of the total medical spending attributable to diabetes (20).

Our results suggest that prescription medication spending has been the primary driver of the growth in per capita diabetes spending. A growing trend of increased spending on prescription medication has also been observed among people without diabetes and general populations in the U.S. and other developed countries (22). However, the increase in prescription medication spending related to diabetes care has well outpaced that of prescription medication spending in the general population (22). From 1999 to 2011, national data suggest that growth in the use and price of prescription medications in the general population is 2.6% and 3.6% per year, respectively; and the growth has decelerated in recent years (22). Our analysis suggests that the growth rates in the use and prices of prescription medications for diabetes patients are considerably higher. The higher rate of growth is likely, in part, due to the growing emphasis on achieving glycemic targets, the use of newer medications, and the use of multidrug treatment strategies in modern diabetes care practice (23,24). In addition, the growth of medication spending is fueled by the rising prices per drug, particularly the drugs that are newly introduced in the market. For example, the prices for newer drug classes such as glitazones, dipeptidyl peptidase-4 inhibitors, and incretins have been 8 to 10 times those of sulfonylureas and 5 to 7 times those of metformin (9).

The increase in excess medical spending on inpatient and ER visits had been entirely driven by the rise in the unit cost, rather than the volume, of medical use. Taking inpatient care as an example, despite the decline in the number of hospital admissions and length of stay among diabetes patients, the average excess inpatient spending due to diabetes increased by $632 (or 65%) between 1987 and 2011. This is merely because, during the same time interval, the unit cost per hospital admission and the unit cost per night of hospital stay for diabetes patients increased by 88% and 273%, respectively. However, it should be noted that after 2001 the increased unit cost attributable to diabetes appeared to slightly decrease, suggesting that the gap of hospitalization costs between diabetes and nondiabetes patients slightly decreased. The growing trend of inpatient costs is consistent with the findings of previous studies (25,26), which had linked the rapid growth in the price of hospitalization to the higher cost of medical devices and supplies. Because of the higher rates of hospitalizations and higher costs per hospitalization, the price growth may have a greater financial impact on diabetes patients and potentially cause them to miss episodes of needed care because of cost.

In absolute terms, the spending growth was ubiquitous in all medical components. However, the contributions of different components have substantially changed. The proportion of prescription medication spending among the total medical spending attributable to diabetes continuously increased, whereas that of inpatient and ER services declined. As a result, prescription medication has now exceeded inpatient services and accounts for the largest component of spending associated with diabetes. Previous studies (23) have suggested that more intensive use of medications improves the control of vascular disease risk factors and thus may reduce inpatient costs. While we did not observe a compensatory reduction in inpatient or ER spending associated with higher drug costs, a declining or plateauing trend for hospital admissions, inpatient stays, and ER visits was observed in this study. This appears to be consistent with the encouraging steady nationwide improvement in diabetes complications (27,28). In turn, it may signify the progress of complication prevention through more aggressive pharmacological treatments, advances in diabetes management care, health system performance, and health promotion efforts for the U.S. diabetes population (29,30).

However, it is also possible that the declining or plateauing trend in the use of inpatient service was driven by other forces. Because this modest decline or plateau was accompanied by a considerable rise in the unit cost of hospitalization during the same period, the reduced use of services may also, in part, reflect circumstances in which diabetes patients forego needed episodes of inpatient or outpatient care because they cannot afford the costs. In addition, since similar trends have also been observed in adults without diabetes, the declining use might merely reflect a national trend that exists in the broad spectrum of medical care.

Our study has several notable limitations. It is possible that the changing health status of the population with diabetes affected our estimates. For example, changing diagnostic thresholds in 1997 and 2003 and enhanced detection during the past 2 decades could have led to the identification of a “diabetic” population that is characterized by lesser degrees of hyperglycemia and better general health. At the same time, this could have been counteracted by decreasing mortality rates, keeping people alive longer with more morbidity. Unfortunately, clinical biomarkers are not available in the data, which prevented us from further examining the effects of the changes. Second, as with other medical spending survey data, respondents in the MEPS and NMES tend to under-report ER and office visits (by 19–30%) (31). It may have led to an underestimation of the spending on the two medical services, and therefore of the total medical spending. However, it is unlikely that it would affect the trend since the underreporting has been found to be consistent across times and surveys (31). Third, the number of prescription medication encounters collected in the MEPS and NMES reflects the quantity of medication purchases, but does not measure the dosage of medication prescribed or the length of supply, which might provide a better description of medication use. Fourth, the medical costs of type 1 and type 2 diabetes are different; however, neither the MEPS nor the NMES allows us to distinguish the types of diabetes. Finally, both the MEPS and NMES surveys are limited to the civilian noninstitutionalized population, and thus the spending for institutional care, such as nursing home care, was not included.

Between 1987 and 2011, medical spending increased both in persons with and in persons without diabetes; and the increase was substantially greater among persons with diabetes. As a result, the medical spending associated with diabetes nearly doubled. The growth was primarily driven by the spending in prescription medications. Further studies are needed to assess the cost-effectiveness of increased spending on drugs. Despite the modest decline in the use of inpatient, ER, and other medical services, the excess medical spending on those services that is attributable to diabetes continued to grow due to a substantial increase in the price of those services.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. X.Z. contributed to the study concept and design and helped to acquire, analyze, and interpret the data; perform the statistical analysis; write the first draft of the manuscript; and perform critical revision of the manuscript for important intellectual content. P.Z. contributed to the study concept and design and helped to analyze and interpret the data, write the first draft of the manuscript, perform critical revision of the manuscript for important intellectual content, and provide administrative, technical, and material support. H.S.K., B.H.B., and R.L. helped to analyze and interpret the data and to perform critical revision of the manuscript for important intellectual content. E.W.G. helped to analyze and interpret the data, critically revise the manuscript for important intellectual content, and provide administrative, technical, and material support. X.Z. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 72nd Scientific Sessions of the American Diabetes Association, Philadelphia, PA, 8–12 June 2012.