Productivity and Medical Costs of Diabetes in a Large Employer Population

The study population was drawn from a Fortune 100 manufacturing firm with facilities throughout the U.S. and >100,000 medical plan beneficiaries, including industrial, service, and professional employees. This company self-insures its health insurance and disability programs. All employees were covered by the health insurance program, and ∼90% were covered by disability benefits. Administrative records include payments made by the employer for medical and prescription drug claims for all beneficiaries (employees, spouses, dependents, and retirees under the age of 65 years) and disability claims (for employees only). The medical and pharmacy data are from fee-for-service claims.

To identify beneficiaries with diabetes in the database, we used a combination of diagnosis codes and drug-specific pharmacy claims. Beneficiaries with diabetes were identified as those who, between 1996 and 1998, had two or more medical and/or disability claims with a diabetes diagnosis code (ICD-9 codes 250.xx) or one or more prescription drug claims for a hypoglycemic agent. A sample of 8,748 beneficiaries with diabetes was identified using this algorithm. Because of the difficulties in differentiating type 1 diabetes patients from those with type 2 diabetes in administrative databases, we elected not to make this distinction in analyses reported here.

We matched each beneficiary with diabetes to one control beneficiary without diabetes over the study period (1996–1998). Subjects with diabetes were matched to control subjects with respect to age (identical numerical age), sex, job classification (salaried/nonsalaried; bargaining/nonbargaining), health plan, and state of residence. Matched control subjects were not selected on the basis of health care or disability claims, and thus the control group included individuals with no claims. When more than one matched control was available, we selected one randomly from among the pool of possible matches using a computerized randomization process in SAS version 8 for Windows (17).

To better understand the reasons for excess illness-related work loss among those with diabetes, we identified comorbidities often associated with diabetes in both the diabetes sample and control subjects. We focused on comorbidities that are recognized in the literature to drive medical costs, morbidity, and mortality in diabetes (2,8,10,18,19,20). These conditions included cardiovascular disease, hypertension, infections related to diabetes (e.g., septicemia, bacteremia), other metabolic diseases (e.g., hyperosmolarity), nephropathy, neuropathy, and retinopathy. Records were searched over the analysis period of the year 1998 for ICD-9 codes related to each comorbidity.

The term “costs” refers to payments by the employer to health care providers or employees for medically related missed workdays. The incremental cost of diabetes for this analysis is the excess cost incurred by diabetes patients over the control subjects. Our goal for matching was to minimize the chance that factors other than diabetes would influence the incremental cost analyses.

Direct costs were reimbursements from the employer to health care providers for inpatient, outpatient, physician, and prescription drug services, as well as for other services (e.g., physical therapy, nursing home services). Costs were reported based on claims for services provided in 1998. Patients’ out-of-pocket and nonmedical costs (e.g., deductibles, transportation) as well as intangible costs are excluded because of the employers’ perspective of the analysis.

Productivity costs in this study are based on “medically related absences,” including both sporadic work loss associated with use of medical services and extended work loss caused by disability. Because actual dates of medical care and disability were known, we counted only workdays (omitting weekends and holidays). Estimates of the cost of time lost from work for medical care were based on the type and frequency of visits to health care providers. If a medical service claim for an employee was recorded during a workweek day, then the following algorithm was applied to determine the number of medically related work absence days: 1) for hospital care, a full day absent for each day in the hospital, and 2) for outpatient or office care, a half day was counted. Because the disability benefit covers lost work beginning with the sixth consecutive day of a work absence, patients who claimed disability were also assigned five medically related work absence days before their period of disability (21). Productivity costs attributable to disability were based on the employer’s payments for disability claims. Productivity costs for medically related sporadic absences were based on the employer’s average wage for the affected employee’s job classification.

Other productivity measures include costs from reductions in work performance while on the job, work loss caused by sporadic absences for illness when medical care was not sought, and friction costs of worker replacement (12,22,23). Our database did not contain information that could be used to estimate these productivity impacts. Hence, our estimates of productivity costs are conservative.

Our aim for the analysis was to assess the average (mean differences) between the diabetes and control populations. Averages are useful to employers or insurers trying to estimate the total expected productivity costs of individuals with diabetes relative to those without diabetes. We calculated prevalence rates of selected comorbidities (e.g., cardiovascular disease, hypertension, and retinopathy) as well as average costs per patient. Medical, productivity, and total (medical plus productivity) costs are reported in 1998 dollars. All cost data are reported as the means ± SD, unless otherwise noted. Descriptive statistics were generated with the SAS version 8 software program (17). We used t tests and χ² tests to evaluate the statistical significance of differences in outcomes of the diabetes sample and matched control subjects.

A sample of 8,170 beneficiaries with diabetes were matched to control subjects without diabetes. The prevalence rate of diabetes was 3.5% in the employer population. This is consistent with estimates for the U.S. that include unemployed and retired individuals (24). Table 1 lists demographic information for the diabetic and matched control populations compared with a 10% random sample of the employer population. The diabetes sample was older (mean age 53 vs. 39 years; P < 0.0001) and more likely to be male (51 vs. 56%; P < 0.0001) than the average beneficiaries. In addition, 19% of the beneficiary population was under 18 years of age in 1998, versus 1% of those with diabetes (P < 0.0001). Similarly, 20% of the beneficiary population was between the ages of 56 and 64 years, versus 47% of those with diabetes. Individuals with diabetes were more likely to be employed than those in the overall beneficiary population (47 vs. 34%; P < 0.001) and less likely to be spouses or dependents (39 vs. 59%; P < 0.001).

The prevalence of diabetes-related comorbidities for diabetic and control subjects is depicted in Fig. 1. Prevalences of all comorbidities were significantly higher in the diabetes sample than in the control group across all age strata (P < 0.0001). In both the diabetes and control groups, cardiovascular disease and hypertension were the most common comorbidities.

On average, beneficiaries with diabetes used more of all types of medical services than control subjects. In 1998, diabetes patients had an average of 3.4 inpatient, 5.5 outpatient, and 9.1 office claims, compared with 1.3, 2.9, and 4.9, respectively, among control subjects. Two factors explain this higher level of service utilization. First, diabetes patients are more likely to use all types of services than control beneficiaries. Of the diabetes patients, 20% used inpatient services, 67% used outpatient services, and 82% used office services, compared with 10, 48, and 62%, respectively, for control subjects. Second, diabetes patients who used medical services had more utilization than control subjects. On average, diabetes inpatient users had 16.6 inpatient claims, outpatient users had 8.0 outpatient claims, and office users had 11.1 office claims, compared with 12.9, 6.2, and 7.8 claims, respectively, for control users of these services.

Table 2 presents the medical and work loss costs for the diabetes sample and matched control subjects. In 1998, the employer’s cost for beneficiaries with diabetes (including employees, their spouses, and dependents) were higher than for beneficiaries without diabetes ($7,778 ± 16,176 vs. $3,367 ± 8,783; P < 0.0001). By subtracting the average costs of the two groups, we find that the incremental medical and productivity cost of beneficiaries with diabetes compared with control subjects was $4,410 ± 18,407. When stratified by age, total medical and productivity costs for beneficiaries in the diabetes sample ranged from $2,589 ± 4,491 for those aged <18 years to $8,568 ± 18,918 for those aged 56–64 years. Costs for beneficiaries in the matched control sample ranged from $1,332 ± 4,750 for those aged <18 years to $3,993 ± 10,542 for those aged 56–64 years. The average total cost for a beneficiary with diabetes ranged from 1.9 to 2.9 times that for a beneficiary in the control population without diabetes, across each age category.

To investigate the sensitivity of the findings to outlier values, we also tabulated median cost values. Table 2 illustrates that the findings are not greatly affected by outliers, because the pattern of the medians is similar to that of the mean values.

As expected, the costs for inpatient care increased with age. This is particularly true among diabetic patients (Table 2). Costs of ambulatory care also follow this pattern and are similar to inpatient costs in magnitude. Each of these components exceeded the prescription drug component in every age category among beneficiaries in the diabetes sample. As expected, productivity costs do not exist in the <18 years of age cohort. Across age cohorts, work loss costs represented a larger proportion of combined costs (range 16.5–31.8%) in the diabetes sample than in control subjects (range 12.8–21.3%). This difference of proportion is associated with higher medical costs among beneficiaries in the diabetes sample. Furthermore, this consideration of work loss costs provides a conservative estimation because they are experienced by employees only, but diluted across all other beneficiaries. Also, work loss costs account for both a larger percentage of total cost and more absolute dollar amounts among the youngest compared with the oldest age group.

Total costs for employees in the diabetes sample were significantly (P < 0.0008) higher than control subjects, ranging from $7,774 (aged 18–35 years) to $10,132 (aged 56–64 years). The incremental medical and productivity cost for employees with diabetes compared with control employees was highest in the younger age cohorts, at $4,671 for those aged 18–35 years, $4,825 for those aged 36–45 years, and $4,369 for those aged 56–64 years. Across all age groups, mean annual work loss costs alone for diabetic employees ranged from 1.7 to 2.2 times that of matched control employees. The proportion of medical and work loss costs attributable to employees in the diabetes sample ranged from 31 to 42%, compared with a range of 33 to 53% for matched control employees. This proportion was larger among control employees because of their lower direct medical costs. Thus, costs for diabetic employees, as a group, are enhanced by the economic burden of medically related work loss.

Across all age groups, the relative likelihood that an employee in the diabetes sample would have at least one disability claim in 1998 for any reason was 33%, versus 20% for an employee in the control group. Among diabetic employees and control subjects who claimed disability during 1998, the mean duration of a disability claim was 41 ± 98 days for diabetic employees, versus 22 ± 73 days for the control group (P < 0.0001). Across age groups, total disability days claimed were significantly higher for diabetic employees. The mean annual number of work loss days for ambulatory visits or hospitalizations was 8 ± 13 days for diabetic employees versus 5 ± 11 days for control subjects.

This study assessed the economic burden of diabetes from an employer’s perspective, considering direct health care costs and productivity costs associated with work loss and disability. Using claims data, we found a mean annual incremental cost of $4,410 for all beneficiaries with diabetes and $4,413 for employees with diabetes compared with control beneficiary and employees, respectively. We found that >30% of the costs associated with diabetic employees are attributable to medically related work absences and disability. Relative to beneficiaries without diabetes, those with diabetes had a larger proportion of costs attributable to work loss, more disability claims, and longer durations of disability. Thus, diabetes represents a significant burden for employers, both in terms of medical and productivity costs. The lower work losses in the oldest group versus the youngest age group may be explained by the “healthy worker” effect (i.e., after a certain age, the sickest individuals drop out of the workforce).

Our diabetes beneficiary sample, although derived from a large employer, appears to be relatively representative of the U.S. population in terms of diabetes prevalence. Also, our findings of expenditures for medical care are consistent with prior studies (5,8,9,10). Several studies have considered the productivity cost burden of diabetes, but they did not use actual employer payments from administrative data (13,14,15,16).

A number of caveats must be considered when interpreting these results. The analysis was limited to administrative claims data and are subject to the usual limitations of such data, such as the absence of clinical detail and validation of its accuracy (12). Because our sample was an employed population, it did not include those who are unemployed, uninsured, or at the upper end of the age range. Therefore, our results do not provide insight regarding unemployed or uninsured patients, or those enrolled in public assistance plans such as Medicaid or Medicare. It may be that, for example, with higher rates of absenteeism or disability, diabetic patients are less likely to be employed. Also, the sample excluded patients enrolled in capitated health plans. Such plans may be expected to enroll healthier beneficiaries. Although the focus of our study was on the work loss costs of morbidity, we did not have the data to consider reduced levels of productivity while at work, incremental sporadic work loss, or the costs of hiring replacement workers. Also, we did not have data to consider the indirect costs of mortality. The analysis also excluded out-of-pocket costs to employees and their families. Although certain classes of expense are unavailable (e.g., child care), an additional issue is that those individuals who do not meet deductibles may still incur substantial costs in the aggregate that will not be captured in the analysis. Moreover, these omitted costs may be higher among the control subjects because their costs will generally be lower. Thus, these results represent a conservative estimate of the economic burden of diabetes to employers.

Our findings that cardiovascular disease and hypertension were much more common among diabetic patients are consistent with other literature. We note that because those with diabetes seek health care services more often than control subjects, additional opportunities exist for diagnosing these comorbid conditions. Furthermore, it may be that the relatively high prevalence of hypertension may also be due in part to lower thresholds for diagnosing hypertension among those with diabetes (25,26).

Working-age individuals with serious illness will cost their employer more in terms of productivity and medical costs than those without serious morbidities. Our study quantifies the economic burden employers face for their employees with diabetes. Given our findings, discriminatory hiring practices and insurance underwriting could pose a threat to working-age individuals with diabetes. In a study using simulated job decisions, diabetic and obese applicants were less likely to be hired because of presumptions of poorer work habits and medically related work absences (27). To address this concern, the federal Health Insurance Portability and Accountability Act of 1996 (HIPAA) provides protection from discrimination in insurance (28). This law prohibits group health insurance plans from using a “preexisting condition” to deny or limit coverage or raise premiums. One implication of our work is that the diabetic community needs to raise its awareness of the potential for employment discrimination and to perhaps consider funding research to identify whether employment discrimination is a serious problem among those with the disease.

Although improved diabetes control has been shown to improve clinical outcomes and reduce medical costs (29,30), it may also lead to a more productive workforce (31). Our results indicate that the burden due to medically related work loss may be substantial for working-age individuals with diabetes. Furthermore, medically related work losses are important in younger as well as older patients. Our study suggests that evaluations of the economic benefits of aggressive diabetes management should not focus solely on direct medical costs, including prescription medications. Indeed, it may be that labor productivity can be enhanced through optimized use of medications (32). We suggest that trials determining the efficacy of intensive therapies to reduce diabetes-related complications should include an evaluation of their impact on productivity for working-age subjects. Controlled studies evaluating the impact of modifications of the work environment that will foster adherence to behaviors known to benefit those with diabetes (ready access to medicines, healthy food choices, and opportunities for exercise) are also warranted.

Several areas of importance related to diabetes and worker productivity are still unknown. More information is needed on how diabetes impacts on-the-job work performance. The relation between poor glycemic control, with its attendant symptoms and productivity, is not well known. Correspondingly, we also need to know more about the impact of improving glycemic control on productivity. Finally, future studies should evaluate the impact of home and workplace-based diabetes management programs on productivity.

This work was performed by Analysis Group/Economics, which was sponsored by an unconditional research grant from Eli Lilly.

The authors would like to thank Pamela Erickson for editorial assistance and Erin Lentz for her data analysis support.