Diabetes itself affected working and wages more than control of blood sugar levels in a Mexican American population.
To determine whether labor market effects were the result of diabetes per se or rather depended on the degree to which diabetes was controlled through management of blood sugar levels.
Methods: This study utilized data from a recently completed survey of households in Brownsville, Texas, a largely Mexican American community with a high prevalence of diabetes that is located on the Texas-Mexico border. Diabetes management, or control, was measured by blood sugar levels, glycosylated hemoglobin (A1C) levels, and interaction terms. Methods used were probit and Heckman regression.
Management of diabetes did not appear to have a discernible impact on labor market outcomes in the short run. However, diabetes was negatively associated with male productivity, particularly in males’ propensity to work. The new American Diabetes Association (ADA) definition of diabetes is based on having an A1C level of <6.5%. Using the new ADA definition, diabetes was negatively associated with female productivity. Female productivity was also negatively associated with the new ADA definition of prediabetes (A1C levels between 5.7% and 6.4%). However, very few people with diabetes actually controlled their blood sugar or A1C levels.
These results imply that in order to avoid productivity losses associated with diabetes, more scarce prevention resources should be spent on prevention of the onset of diabetes than on the management of A1C for those already diagnosed with diabetes. For women, the prevention of prediabetes is key.
(Am J Manag Care. 2011;17(8):569-576)
Utilizing data from a largely Mexican American community with a high prevalence of diabetes, we tested whether labor market effects were the result of diabetes per se or depended on the degree to which diabetes was controlled through management of blood sugar levels. We observed that:
Diabetes is a chronic health condition which has substantial economic implications. The American Diabetes Association (ADA) estimated that diabetes cost the US economy $174 billion in 2007.1 An important part of this cost is associated with lower labor productivity.2-6 Public health officials argue that diabetes prevention is important, but prevention could mean 1 of 2 things: prevention or delay of the onset of diabetes, or prevention of diabetes-related problems through the management or control of glycosylated hemoglobin (A1C) levels. If the productivity costs of controlled diabetes are low, scarce prevention dollars could be concentrated on the much smaller population already diagnosed with diabetes. On the other hand, if the costs associated with diabetes are substantial whether the condition is managed or not, then prevention dollars must be spread over the much larger general population. There is currently no information to inform policy makers how to apportion scarce prevention dollars between diabetes onset and diabetes management.
A clue to untangling whether labor productivity is due to diabetes per se or uncontrolled diabetes lies in the differential effects of diabetes on male and female labor productivity. While the labor costs of diabetes are clear for men, the effect of diabetes on female productivity is less clear.3,4 It is hypothesized, but has not been proved, that women are more inclined to manage their diabetes than men and therefore face a lower labor productivity penalty due to diabetes.4 Management of blood sugar levels can occur through a combination of physician visits, diligent use of pharmaceuticals, use of monitoring devices, and health behavior modification. Many national surveys (eg, the National Health Interview Survey) have self-reported indicators of physician visits, exercise, and diet. Our study features the gold standard summary variable for health behaviors: individual, laboratory-measured A1C levels as well as blood sugar levels. Our results enabled us to determine whether women manage their diabetes better than men and, if so, whether this better self-management partly or fully explains the differential effects in terms of labor productivity.
While overall diabetes-related labor costs are rising, it is not clear that labor costs associated with diabetes are increasing at the individual level. If per capita diabetes-related productivity costs are falling, a portion of prevention dollars should be shifted toward managing, rather than preventing, diabetes. Technological changes over the last 3 decades have led to changes in the labor market and in the medical field. First, there has been an increase in the number of jobs that are less physically demanding and therefore accessible to persons who have diabetes or who are obese.4 Second, new drugs, A1C monitoring devices, and food science advances (eg, artificial sweeteners) are making diabetes management easier and less costly than ever before. On the other hand, environmental changes (eg, fast food availability) continue to make preventing or delaying the onset of diabetes more difficult.7 However, these environmental changes also make managing diabetes more difficult.
This study adds 2 important elements to the growing literature on diabetes and labor market outcomes. First, we examined whether poor diabetes management was the cause of adverse labor market outcomes rather than diabetes per se. In our study, A1C levels were measured in a laboratory for all participants, whether or not they were diagnosed with diabetes by a physician. Second, we examined whether diabetes management explains the lack of association between female productivity and diabetes.
Our study population, Mexican Americans living on the border of Mexico, had a high prevalence of diabetes. The percentage of this population diagnosed with this health condition is expected to rise from 1.4 million nationally in 2002 to 2.9 million in 2020, a 107% increase.8 In contrast, the total US population diagnosed with diabetes is expected to increase by 44% during the same time period.
DIABETES SELF-MANAGEMENT AND HEALTH CAPITAL
Glycosylated hemoglobin measures blood glucose over a 2- to 3-month period. The greater the percentage of sugar that is in the blood supply, the higher the percentage of hemoglobin that has been glycosylated. Unlike blood sugar readings, the percentage of hemoglobin that is glycosylated is invariant to day-to-day blood sugar levels.9 If one’s management (eg, exercise, diet) is consistent, A1C levels will be better than if management is consistently poor. Thus, A1C is an excellent measure of health capital or stock in the Grossman sense.10 Blood sugar, which we will also consider, varies more because it depends on recent behavior,9 and therefore is less reliable as a health stock measure.
For persons with diabetes, having A1C levels of 7% or less indicates that the person is managing his or her diabetes. 11 (The ADA recommends that diabetes be diagnosed using an A1C level of 6.5% or more.11) Labor productivity for adults with diabetes who have low A1C levels should be higher than for persons who have high
A1C levels, even if A1C levels are not below 7%.1 On the other hand, an alternative hypothesis is that time and effort expended to manage diabetes may come at the expense of current, short-term labor productivity. Thus, it may be that efforts to manage diabetes lower productivity in the short term.
Our strategy to test the hypothesized association between labor productivity and unmanaged diabetes involved creating an interaction term between diabetes and A1C levels. Persons with diabetes had an interaction term value of 7 or lower if they were managing their diabetes, but persons with diabetes who were not managing their diabetes had an interaction term value of greater than 7. Persons without diabetes had an interaction term value of zero. Thus, we expected to find a negative relationship between this interaction term and log wage. We also examined the association between blood sugar levels interacted with diabetes status (persons who managed their diabetes would tend to have lower blood sugar levels) and employment and wages.
DATA AND METHODS
The data for this study came from a household survey conducted in Brownsville, Texas, a metropolitan area with a total population of 175,494 in 2008, located in the US-Mexico border region.12 Brownsville is the most southernmost city in Texas and 91.3% of its population is of Hispanic origin.13 Brownsville, like many other Texas border cities, is characterized by high poverty levels and low educational attainment. For example, 36.5% of people in Cameron County, where Brownsville is located, live below the poverty level (which ranks the county next to last in the United States). Only 60% of the population aged 25 years and over have completed high school, which ranks the county last in the United States. Approximately 46.5% of children under age 18 years are below the poverty level, which ranks the county next to last in the United States.14
In order to ensure variation in household income, the survey team selected census blocks from census tracts with median household incomes in the first and the third quartiles. (Note that households selected within the census tracts may have differing income levels.) The 2000 US Census of Population and Housing was used to select the probability sampling frame. Within all randomly selected clusters (census blocks), all the households in the census block were contacted. A participant between ages 35 and 64 years from each household was randomly selected using a 10-digit permutation algorithm.
For our purposes, the key features of our data were the laboratory-measured A1C levels, blood sugar levels, and labor variables. To our knowledge, the Health and Retirement Study is the only national data set that features A1C as a variable, but the average age in that sample is much older. Our sample was exclusively Mexican in origin, so the diabetes prevalence was higher despite it being a younger sample than that in the Health and Retirement Study. The study was approved by the University of Texas—Houston Committee for the Protection of Human Subjects.
Probit and Heckman Models
We first estimated probit models for working status si (1 = yes, at least part-time, 0 = no)
(1) si = zi Y v,
where zi is a vector of characteristics related to working, including human capital variables, diabetes status, diabetes management, and so forth, and Y represents associated coefficients. v is a random error term.
In order to account for sample selection bias, we also estimated the log wage yi equation using the standard Heckman method using full maximum-likelihood.15 Following Baum,15
(2) yi = xi B u
(3) si = I(zi Y v > 0).
The Heckman model solves the selection problem by estimating the probability that wage is observed si (1 = yes, at least part-time, 0 = no) in equation 3. In equation 2, xi is a vector of characteristics related to working, including human capital variables, diabetes status, diabetes management, and so forth, and B represents associated coefficients. The function I(·) returns 1 when the condition is true, 0 otherwise. zi, Y, and v are defined in equation 1. Note that accounting for selection is necessary when the correlation of u and v is nonzero. We included the log of other household income in the selection equation, but excluded it in the log wage equation in order to ensure identification.
Productivity was measured by log wage (yi) in equation 2. The idea is that after controlling for human capital variables, differences in wages for people with similar educational and experience levels are due to differences in productivity, which may partly be explained by health differences. Work propensity, or the likelihood of working versus not working, is accounted for in equation 3.
shows the laboratory data for survey participants with diabetes. Despite the general perception that women are more compliant with diabetes management, women had 10% higher A1C levels than men. This observation casts doubt on the notion that better labor outcomes for women are the result of better diabetes management.4 For both sexes, the confidence intervals, which were weighted to the population with diabetes, were narrow. Neither men nor women appeared to be managing their diabetes effectively.
reports the descriptive statistics for the variables used in the probit and Heckman regressions by sex. Note that women were less likely to be employed and had slightly less schooling than men, and were more likely to be immigrants. Men also were paid more than women. For both men and women, the sample size was weighted to the population size (208/7686 men; 436/9762 women).
shows the results of the probit estimates by sex. These results were weighted to the population level of the first and third quartiles in Brownsville, Texas. Most signs were as expected, but were not statistically significant. For men, one would expect years residing in Brownsville and years of schooling to be positively related to the likelihood of working, but they were not significantly greater than zero. For most respondents to this survey, the number of years residing in Brownsville was synonymous with the number of years in the United States.
For men, having diabetes lowered the likelihood of working by 5%. For women, the relationship between diabetes and working was positive, but not statistically significant. These results are similar to others3,4 in that among persons with diabetes, men experienced greater productivity losses than women. Note further that for Hispanic men, diabetes has been shown to be endogenous. However, the primary aim of this article was to examine the direct effect of managing diabetes on productivity. We did not have any suitable instruments to examine diabetes management behavior.
shows the effects of blood sugar levels and A1C, respectively, on the likelihood of working. The controls were the same as in Table 3. Neither blood sugar nor A1C significantly affected the likelihood of working, and for women, the signs were not as expected. However, that is in accordance with previous research showing that diabetes affects men more than women.3,4
shows the effects of blood sugar levels and A1C, interacted with diabetes, on the likelihood of working, with the controls the same as in Table 3. The idea behind the interaction, as described earlier, is that nondiabetic persons should not have their work propensity affected by lower blood sugar or A1C levels. However, for those with diabetes, managing blood sugar and A1C may increase work propensity. For men, both the blood sugar interaction and the A1C interaction were significantly (negatively) related to work propensity. Glycosylated hemoglobin was more tightly correlated with work propensity than were blood sugar levels, likely because A1C is a more stable measure of diabetes management. If the A1C level for an adult with diabetes fell by 10% due to better diabetes management, the likelihood of working increased by 0.44%. However, note that having diabetes decreased the likelihood of working by 5%. For women, the signs were not as expected, but the elasticities were not significantly different from zero.
reports the estimates for the Heckman model. For the selection model in the lower half of Table 6, the signs of the coefficients were as expected for both sexes, except for schooling among males. Being born in Mexico had a positive effect on working propensity for men, but not women. The variable “years in Brownsville” was negatively related to the likelihood of working. The selection model was warranted because both likelihood ratio tests rejected the null hypothesis that the selection equation and log wage equations had independent error terms.
The top half of Table 6 reports the log wage estimates for the Heckman model. The signs of the coefficients were as expected for both sexes, except not all were significant.
Diabetes was negatively related to log wages. The wage premium was modest—having diabetes lowered wages by 0.74% for men, but there was no significant effect for women. The result mirrors that of other studies in that wage penalties were restricted to men.3,4
shows the results of Heckman models estimated for the laboratory measures, blood sugar, and A1C levels. The other independent variables are as in Table 6. For women only, the effect of blood sugar levels was negatively related to log wage at the 10% level. However, the marginal effect of controlling blood sugar levels on log wage was small—a 10% reduction in blood sugar levels only raised log wages by 0.87%.
The results in Table 7 for A1C levels are similar, except that A1C levels had small effects on log wages for women. The effects of A1C on log wage were higher than those for blood sugar—a 10% reduction in blood sugar levels raised log wages by 1.2%. The association between blood sugar and wages was lower than the association between A1C and wages, perhaps due to A1C being a more stable measure of health stock.
shows the results of Heckman models estimated for the laboratory measures, blood sugar, and A1C levels interacted with diabetes. For persons without diabetes, variation in A1C or blood sugar levels should not matter for productivity, and with interaction terms, these variables were set to zero. The other independent variables were again as shown in Table 6. For men, the effect of blood sugar interacted with diabetes was negatively related to log wage and to selection into employment. However, the marginal effect of controlling blood sugar levels on log wage was small as with blood sugar— a 10% reduction in blood sugar levels only raised log wages by 0.57%. For women, there was no significant effect.
The results in Table 8 for A1C levels interacted with diabetes were similar to those for A1C levels, except that the A1C interaction had larger effects for men and very small, but positive, effects for women. For men, a 10% reduction in blood sugar levels raised log wages by 0.62%.
NEW DEFINITION OF DIABETES
In 2010, the ADA began recommending diagnosis of diabetesbased on having an A1C level of greater than 6.5%, a recommendation it had avoided before because measurement methods were previously not standardized.11 They also recommended defining persons as having prediabetes (ie, at risk for diabetes) if their A1C levels were between 5.7% and 6.4%.11 Our results show that threshold levels of A1C were associated with productivity losses for males. A natural question arises as to whether thresholds for productivity effects also differed by A1C. We saw in Table 7 that the log wage for women was associated with A1C, but not the log wage for men. Table 8 showed an association between managed diabetes (A1C levels) and log wage for men, but not for women.
Figure 1 shows the elasticities of Heckman estimates for various hypothetical definitions based on A1C threshold levels. For instance, the left-most point on the horizontal axis shows the elasticity of diabetes on labor productivity based on having an A1C level greater than 5%, the next point shows the elasticity of diabetes on labor productivity based on having an A1C level greater than 6%, and so on. Neither male nor female productivity was affected by having an A1C level of 5% or more; prediabetes was associated with productivity losses in women, but not men.11 Male productivity was not associated with an A1C level of 7% or more, but female productivity was. At A1C threshold levels of 8%, 9%, 10%, and 11%, the interaction terms were negatively associated with male productivity, but not female productivity.
repeats the analysis shown in Figure 1, except that the hypothetical diabetes cases based on A1C are interacted with A1C itself. The results are similar to those shown in Figure 1.
DISCUSSION AND CONCLUSION
Our results show that diabetes, whether managed or not, lowers male work propensity by approximately 5 percentage points. For males, the wage productivity premium for avoiding diabetes was approximately 0.74%. When diabetes management improved A1C levels by 10%, the likelihood of working increased by 0.44% and wages rose by 0.62%; the average male in our sample would need to drop 2 percentage points of A1C for his diabetes to be considered managed, which would increase the likelihood of working by nearly 1 percentage point and raise wages by 1.4%. Thus, diabetes perse is the stronger driver of productivity losses than the level of diabetes control. However, A1C levels were consistently more strongly negatively associated with productivity than blood sugar levels.
Using extant clinical definitions of diabetes, we saw little evidence of diabetes, whether controlled or not, affecting female productivity. However, female productivity was negatively associated with new 2010 ADA definitions of diabetes and prediabetes,11 whereas male productivity was not. The average female in our sample would need to drop 3.1 percentage points (or 31%) of A1C for her diabetes to be considered managed, which would raise wages by 0.59%.
The implications of these results are that in order to avoid productivity losses associated with diabetes, more scarce prevention resources should be spent on prevention of the onset of diabetes than the management of A1C for those already diagnosed with diabetes. Unfortunately, that means resources must be spread more thin rather than concentrated on the smaller group with diabetes. On the other hand, we had very few examples of people with diabetes who actually were controlling their blood sugar or A1C levels, so we could not determine whether fully controlled diabetes would improve productivity, which is a limitation of the study. Worse news still is that attention must be paid to preventing prediabetes, especially in the case of women.
Even though it is likely that productivity in the future will be higher for adults with diabetes who currently are managing their health condition, management takes time in the short run. Persons must exercise, change their diet, consult with a physician, check monitors, and so forth—all of which add to health stock in the Grossman sense,10 but the opportunity cost of these actions may detract from productivity in the short run.
Our results are consistent with previous research suggesting that diabetes mainly affects the labor market of men, but not women.2,3 However, our results do not support the conjecture made by Kahn4 that better management of diabetes by females is the reason that labor market outcomes are better.
Our results on self-management of diabetes are valid in the short term only. The Grossman model suggests that health capital leads to long-run productivity gains. However, we only compared short-term effects of diabetes management with crosssectional data. The number of female participants was higher than the number of male participants. Finally, as noted earlier, very few people with diabetes successfully managed their blood sugar levels.
Author Affiliations: From University of Texas School of Public Health (HSB, AP, CLH, SPF-H, JBM), Brownsville; University of Texas at Austin (LMY); Department of Health Management and Policy (JAP), University of North Texas Health Science Center, Fort Worth; Leonard Davis Institute of Health Economics (JAP), University of Pennsylvania, Philadelphia.
Funding Source: Financial support for this study was provided by the Hispanic Health Research Center at the Brownsville Regional Campus of the University of Texas School of Public Health (NIH CMHD P20 MD000170-04).
Author Disclosures: The authors (HSB, AP, LMY, JAP, CLH, SPF-H, JBM) report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.
Authorship Information: Concept and design (HSB, AP, JAP, CLH, SPF-H, JBM); acquisition of data (HSB, AP, CLH, SPF-H, JBM); analysis and interpretation of data (HSB, AP, LMY, SPF-H); drafting of the manuscript (HSB, AP, LMY, JAP); critical revision of the manuscript for important intellectual content (HSB, AP, JAP); statistical analysis (HSB, AP, LMY); provision of study materials or patients (HSB, SPF-H, JBM); obtaining funding (HSB, AP, CLH, SPF-H, JBM); administrative, technical, or logistic support (HSB, AP); and supervision (HSB, AP).
Address correspondence to: H. Shelton Brown III, PhD, Division of Management, Policy and Community Health, Michael & Susan Dell Center for Advancement of Healthy Living Institute for Health Policy, University of Texas School of Public Health-Austin Campus, University of Texas Administration Building (UTA), 1616 Guadalupe Street, Suite 6.300, Austin, TX 78701. E-mail: firstname.lastname@example.org.
1. American Diabetes Association. Economic costs of diabetes in the U.S. in 2007 [published correction appears in Diabetes Care. 2008; 31(6):1271]. Diabetes Care. 2008;31(3):596-615.
2. Bastida E, Pagán JA. The impact of diabetes on adult employment and earnings of Mexican Americans: findings from a community based study. Health Econ. 2002;11(5):403-413.
3. Brown HS 3rd, Pagán JA, Bastida E. The impact of diabetes on employment: genetic IVs in a bivariate probit. Health Econ. 2005;14(5): 537-544.
4. Kahn ME. Health and labor market performance: the case of diabetes. J Labor Econ. 1998;16(4):878-899.
5. Ng YC, Jacobs P, Johnson JA. Productivity losses associated with diabetes in the US. Diabetes Care. 2001;24(2):257-261.
6. Lavigne JE, Phelps CE, Mushlin A, Lednar WM. Reductions in individual work productivity associated with type 2 diabetes mellitus. Pharmacoeconomics. 2003;21(15):1123-1134.
7. Powell L, Bao Y. Food prices, access to food outlets and child weight outcomes. Econ Hum Biol. 2009;7(1):64-72.
8. Hogan P, Dall T, Nikolov P; American Diabetes Association. Economic costs of diabetes in the US in 2002. Diabetes Care. 2003;26(3):917-932.
9. Woerle HJ, Pimenta WP, Meyer C, et al. Diagnostic and therapeutic implications of relationships between fasting, 2-hour postchallenge plasma glucose and hemoglobin a1c values. Arch Intern Med. 2004; 164(15):1627-1632.
10. Grossman M. The Demand for Health: A Theoretical and Empirical Investigation. New York: National Bureau for Economic Research, distributed by Columbia University Press; 1972.
11. American Diabetes Association. Standards of medical care in diabetes— 2010 [published correction appears in Diabetes Care. 2010;33(3): 69]. Diabetes Care. 2010;33(suppl 1):S11-S61.
12. US Census Bureau. American Factfinder. Technical report. Generated by Shelton Brown. http://factfinder.census.gov/home/saff/main.html?_lang=en. Published 2008.
13. US Census Bureau. American Factfinder. Technical report. Generated by Shelton Brown. http://factfinder.census.gov/home/saff/main.html?_lang=en. Published 2006.
14. US Census Bureau. American Factfinder. 2003 American Community Survey. http://factfinder.census.gov/servlet/DatasetMainPageServlet?_ program=ACS& _submenuId=datasets_2&_lang=en. Accessed July 7, 2011.
15. Baum CF. An Introduction to Modern Econometrics Using Stata. College Station, TX: STATA Press; 2006.