Skip Navigation


American Journal of Epidemiology Advance Access originally published online on December 22, 2005
American Journal of Epidemiology 2006 163(4):342-351; doi:10.1093/aje/kwj027
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
163/4/342    most recent
kwj027v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (4)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Port, S. C.
Right arrow Articles by Goodarzi, M. O.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Port, S. C.
Right arrow Articles by Goodarzi, M. O.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

American Journal of Epidemiology Copyright © 2005 by the Johns Hopkins Bloomberg School of Public Health All rights reserved; printed in U.S.A.

Original Contribution

The Predictive Role of Blood Glucose for Mortality in Subjects with Cardiovascular Disease

Sidney C. Port1,2, Noel G. Boyle3, Willa A. Hsueh4, Manuel J. Quiñones4, Robert I. Jennrich2 and Mark O. Goodarzi5

1 Department of Mathematics, University of California Los Angeles, Los Angeles, CA
2 Department of Statistics, University of California Los Angeles, Los Angeles, CA
3 Division of Cardiology, UCLA Medical Center, Los Angeles, CA
4 Division of Endocrinology, Diabetes, and Hypertension, David Geffen School of Medicine, University of California, Los Angeles, Los Angeles, CA
5 Division of Endocrinology, Diabetes and Metabolism, Cedars-Sinai Medical Center, Los Angeles, CA

Correspondence to Dr. Sidney C. Port, Department of Mathematics, University of California Los Angeles, Los Angeles, CA 90095-1555 (e-mail: sport{at}ucla.edu).

Received for publication April 2, 2005. Accepted for publication September 2, 2005.


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
 References
 
Using the Framingham Heart Study data (United States, 1948–1978), the authors examined the association of blood glucose with 2-year all-cause, cardiovascular, and noncardiovascular mortality in subjects with documented cardiovascular disease. After adjustment for systolic blood pressure, cholesterol, body mass index, cigarette smoking, and use of antihypertensive agents, they found that glucose was a strong, independent predictor of mortality. However, the relations for men and women were qualitatively different. For men, adjusted mortality risk increased very rapidly through the normal range (from 4.12% at 3.89 mmol/liter (70 mg/dl) to 12.26% at 5.55 mmol/liter (100 mg/dl)) and was flat at 12.26% thereafter. For women, risk was flat at 3.65% through the normal range and then increased rapidly, reaching 8.34% at 6.99 mmol/liter (126 mg/d), but increased much more slowly thereafter. Exactly analogous relations held for cardiovascular mortality. For men and women combined, noncardiovascular mortality increased from 1.82% at 3.89 mmol/liter to 2.06% at 5.55 mmol/liter to 2.29% at 6.99 mmol/liter (p for trend = 0.009). These findings suggest that although 5.55 mmol/liter (normal) may be a useful mortality risk division (albeit with different implications for the two sexes), 6.99 mmol/liter (diabetic) is not, especially for men.

blood glucose; cardiovascular diseases; mortality; risk factors

Abbreviations: AIC, Akaike's Information Criterion; CVD, cardiovascular disease


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
 References
 
A growing body of evidence suggests that glucose may be an important predictor of mortality in patients with established cardiovascular disease (CVD) (1Go–7Go). Above-normal glucose levels below the diabetes cutpoint have been shown to carry increased mortality risk for patients in an acute phase of coronary heart disease (1Go–3Go) or undergoing percutaneous coronary intervention (5Go). Only a few studies have examined the role of glucose level in mortality among subjects in a chronic, stable phase of CVD (4Go, 6Go, 7Go). Some indicated that risk increased across the three broad categories: normal (≤5.55 mmol/liter (100 mg/dl)), impaired (5.56–6.99 mmol/liter (101–126 mg/dl)), and diabetic (>6.99 mmol/liter (>126 mg/dl)) (4Go, 6Go). Recently, using the Framingham Heart Study data, Port et al. (7Go) showed that there was a continuous, graded relation of all-cause, cardiovascular, and noncardiovascular mortality to blood glucose for nondiabetic subjects with chronic CVD. These studies considered only men and women combined. In the present study, we used the same Framingham data as those used by Port et al. to quantitatively examine the mortality/glucose relations over the entire glucose range for men and women separately as well as together. Our primary goals were to determine 1) whether the relations differed for men and women and 2) whether the current normal (5.55 mmol/liter (100 mg/dl)) and diabetic (6.99 mmol/liter (126 mg/dl)) cutpoints are useful divisions for mortality risk.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 APPENDIX
 References
 
Background
The methodology of the Framingham Heart Study has been described previously (8Go). In brief, the target population consisted primarily of White, urban, middle-class Americans. In 1948, a sample of 5,209 subjects was selected from Framingham, Massachusetts; these subjects were given biennial examinations. The 30-year data used here contain the results for the first 15 examination cycles. The Framingham Heart Study determined blood glucose values using a sample of random (i.e., casual) whole blood using Nelson's method (9Go). For ease of comparison with other studies, here we state all of our results in terms of plasma equivalents (glucose) obtained by applying the following formula: plasma glucose = 1.12 x whole blood glucose (10Go). Glucose was not measured at examinations 5, 11, and 13.

According to Framingham classifications, CVD consisted of coronary heart disease, stroke, congestive heart failure, and intermittent claudication. Coronary heart disease consisted of myocardial infarction, coronary insufficiency, sudden death, and angina pectoris. CVD status was ascertained by direct examination by two or more Framingham physicians using a variety of techniques, as detailed by Dawber (8Go).

Statistical analysis
Most prospective studies, having just single initial measurements, must look long term to accrue sufficient events. However, by using the Framingham investigators' pooled logistic regression method "PRO" (pooling of repeated observations) (11Go, 12Go), we took full advantage of the fact that we had 15 repeated measurements of baseline variables and used all deaths that accrued over 30 years of follow-up to accurately determine 2-year risks. That method, which has close affinities with survival analysis with time-varying covariates, considers each examination as a separate ministudy and pools the observations to follow subjects anew over successive 2-year intervals until death or the end of the 30-year follow-up. We considered the pooled population of subjects who, at the time of the examination, were between ages 45 and 74 years and had a current or previous diagnosis of CVD. This pooled population was divided into six age-sex groups: 1) men aged 45–54 years, 2) men aged 55–64 years, 3) men aged 65–74 years, 4) women aged 45–54 years, 5) women aged 55–64 years, and 6) women aged 65–74 years.

All models were fitted by logistic regression using Stata 9 software (Stata Corporation, College Station, Texas). Age and sex were entered categorically, and all models were adjusted for age-sex category, systolic blood pressure, total cholesterol, body mass index, cigarette smoking, use of antihypertensive drugs, and time period (i.e., examination). Significance of terms was decided by likelihood ratio and two-sided Wald tests.

We first examined risk associated with the three broad glucose categories: normal (≤5.55 mmol/liter (100 mg/dl)), impaired (5.56–6.99 mmol/liter (101–126 mg/dl)), and diabetic (>6.99 mmol/liter (>126 mg/dl)) for men and women separately as well as combined. We next studied risk by considering glucose as a continuous covariate. Our basic structure for continuous models was logit = ßx + f(g), where x is a vector of covariates used for adjustments, and f is a specified (up to unknown parameters) function of glucose g. We examined the mortality/glucose relation with minimal assumptions on the shape of that relation by using the "nine-categories" model that assumed risk to be constant only within each of nine glucose categories (mmol/liter (mg/dl)): ≤3.89 (≤70), 3.90–4.44 (71–80), 4.45–5.00 (81–90), 5.01–5.55 (91–100), 5.56–6.11 (101–110), 6.12–6.99 (111–126), 7.00–9.71 (127–175), 9.72–12.49 (176–225), and >12.49 (>225). Specifically,

Formula
That model, which was closer to the actual data than low-parameter continuous fits, was used primarily to visualize the glucose/mortality relation in more detail than provided by using just normal, impaired, and diabetic categories, and to aid in judging how well low-parameter continuous models reflected the actual data.

The Appendix gives details of the low-parameter models we considered. Included were the "linear" model [f(g) = {alpha} + {gamma}g]; transform models with f(g) = {alpha} + {gamma}h(g), where h is a function of g; linear-transform models with both g and h(g) as predictors [f(g) = {alpha} + {gamma}1g + {gamma}2h(g)]; and, importantly, spline models with possible knots at k1 = 5.55 (current normal), k2 = 6.11 (previous normal), and k3 = 6.99 mmol/liter (diabetic) cutpoints. These knots were chosen a priori to evaluate whether they were potentially useful as cutpoints for mortality. Low-parameter models were compared by using Akaike's Information Criterion (AIC) (13Go).

We examined the adequacy of the low-parameter fits by comparing the rates they predicted in the normal, impaired, and diabetic categories, and in the nine categories, with the adjusted observed rates in these categories. To facilitate visual comparison of the nine-categories model with low-parameter models, we plotted the nine-parameter model as a step function whose height was the risk in a given category and whose horizontal segments marked the location of the observations in a category; we singled out, on each segment, the location of the mean glucose level in the category.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 APPENDIX
 References
 
Overall, for men and for women, age, systolic blood pressure, body mass index, and percentage of antihypertensive drug users tended to increase with increasing glucose level. Women had higher levels of all risk factors that we adjusted for other than smoking (table 1). There were 3,836 observations (57 percent men) and 310 deaths (70 percent men), of which 69.7 percent (n = 216) were cardiovascular related (table 2).


View this table:
[in this window]
[in a new window]
  		TABLE 1. Pooled population characteristics by sex and by normal, impaired, and diabetic glucose categories, Framingham Heart Study, 1948–1978

 

View this table:
[in this window]
[in a new window]
  		TABLE 2. Deaths by sex and by normal, impaired, and diabetic glucose categories,* Framingham Heart Study, 1948–1978

 
The Framingham Heart Study classified a subject as "glucose intolerant" at a given examination if any of the following conditions held: 1) previous or current diagnosis of diabetes, 2) definite or trace glucose in urine, and 3) plasma glucose ≥7.46 mmol/liter (134 mg/dl). A total of 15.6 percent of women and 17.4 percent of men were known to be glucose intolerant. Essentially none of the subjects were using lipid-lowering agents, and none of the non-glucose-intolerant subjects were using antihyperglycemic agents.

All-cause mortality
Men and women combined.
For men and women combined, adjusted all-cause mortality rates significantly increased across the normal, impaired, and diabetic categories (figure 1). The nine-categories model suggested that risk rapidly increased through the normal or subdiabetic range and increased much more slowly thereafter (figure 2). The low-parameter fits supported this view. Single-knot spline models with knots at 5.55 or 6.11 mmol/liter, or the two-knot spline with these knots and the right slope constrained to be 0, were AIC equivalent, optimal (appendix table 1), and predicted risks that were very close to each other (data not shown). We selected the spline with knot at 5.55 mmol/liter as our "all-subjects spline" model. That model was AIC superior to the linear model (appendix table 1). In the all-subjects spline model, both the left slope and the right slope were significant (p < 0.001 and p = 0.050, respectively), and there was a significant difference between them (p = 0.001): the left slope was 12.02 times greater than the right slope (figure 2). However, as detailed below, the relation for the sexes combined presented a biased picture of the relation for men and women individually.


Figure 1
View larger version (9K):
[in this window]
[in a new window]
  		FIGURE 1. Adjusted 2-year death rates/100 for men and women combined, by normal, impaired, and diabetic glucose categories, Framingham Heart Study, 1948–1978. (Refer to the Materials and Methods section of the text for adjustment details.) The black bars give the adjusted all-cause mortality rates in each category, together with their 95% confidence intervals. The p values are Holm adjusted for multiple testing. The white bars show the rates predicted in these categories from an optimal spline model. The rates predicted in the three glucose categories by the spline model were quite close to those observed.

 

Figure 2
View larger version (9K):
[in this window]
[in a new window]
  		FIGURE 2. Adjusted 2-year rates of death from all causes for men and women combined, by glucose level, predicted by three models, Framingham Heart Study, 1948–1978. Linear model (dashed curve); spline model with knot at 5.55 mmol/liter (solid curve). The horizontal dashed intervals are fits from the nine-categories model, the segments of which mark points where there is at least one observation; the black triangles on each interval mark the location of mean glucose in that category. (Refer to the Materials and Methods section of the text for further information.) The spline model provides a considerably better explanation of the mortality risk than the linear model and shows that mortality rates rise quickly over the normal range and then rise much more slowly thereafter.

 
Men and women separately.
Surprisingly, there was a qualitative difference between the sexes in the association of glucose with mortality (figure 3). For men, there were significant differences in mortality between the normal and impaired and between the normal and diabetic categories, but no difference was found between the impaired and diabetic categories. For women, there was a large and significant difference between the impaired and diabetic categories and only a small (nonsignificant) difference between the normal and impaired categories.


Figure 3
View larger version (18K):
[in this window]
[in a new window]
  		FIGURE 3. Adjusted 2-year death rates/100 for men (upper panel) and women (lower panel) separately, by normal, impaired, and diabetic glucose categories, Framingham Heart Study, 1948–1978. (Refer to the Materials and Methods section of the text for adjustment details). The black bars give the adjusted all-cause mortality rates in each category, together with their 95% confidence intervals. The p values are Holm adjusted for multiple testing. The white bars show the rates predicted in these categories from an optimal spline model. Comparison with figure 1 shows that the relation for men and women combined is considerably biased when applied to each sex individually.

 
Examining the relation with glucose as a continuous covariate revealed much more dramatic differences (figure 4). The nine-categories model suggested that for men, risk rose very rapidly through the normal range and then leveled off, while, remarkably, the reverse seemed true for women. For them, risk seemed unrelated to glucose until about the top of the normal range and then increased. The low-parameter models supported these impressions.


Figure 4
View larger version (16K):
[in this window]
[in a new window]
  		FIGURE 4. Adjusted 2-year rates of death from all causes for men (upper panel) and women (lower panel) separately, by glucose level, predicted by three models, Framingham Heart Study, 1948–1978. Linear model (dashed curve); optimal spline models (solid curve). The horizontal dashed intervals are fits from the nine-categories model, the segments of which mark points where there is at least one observation; the black triangles on each interval mark the location of mean glucose in that category. (Refer to the Materials and Methods section of the text for further information.) Compared with the spline fits, the linear fits are poor, especially for men. The relation of mortality to glucose is very different for the two sexes. For men, risk rises extremely rapidly through the normal range and is constant thereafter. For women, risk is constant through the normal range, rises rapidly through the impaired range, and then rises much more slowly thereafter. Comparison with figure 3 shows that the relation for men and women combined is very biased when applied to each sex individually.

 
For men, the clear optimal AIC model was the single-knot spline with knot at 5.55 mmol/liter and right slope constrained to be 0. We chose that model as our "men-spline" model. It had a large, significant left slope (p < 0.001) and was vastly superior to the linear model (appendix table 1 and figure 3). According to the men-spline model, risk increased very rapidly from 4.12 percent at 3.89 mmol/liter (70 mg/dl) to 12.26 percent at 5.55 mmol/liter (100 mg/dl) and was flat at 12.26 percent thereafter (figure 3). The unconstrained spline model with knot at 5.55 mmol/liter supported the flatness to the right; in that model, there was a small, nonsignificant (p = 0.889), negative right slope.

For women, the two-knot spline models with left slopes equal to 0 were equivalent AIC optimal models, were AIC superior to the linear model (appendix table 1), and provided risk predictions that were very close to each other (data not shown). The various equivalent possibilities for women pointed to a relation in which risk was unrelated to glucose until some point between 5.55 and 6.99 mmol/liter; then, risk first increased very rapidly and then increased at a much diminished rate thereafter. The equivalences also indicated that data were insufficient to further refine the shape of the relation in the above-normal range beyond asserting that risk increase rate in the impaired range was significantly greater than in the diabetes range or to further isolate at what point the increase in risk between 5.55 and 6.99 mmol/liter began.

We selected the spline with knots at 5.55 and 6.99 mmol/liter as our "women-spline" model, which conservatively allowed risk increase to begin at the lowest value and rapid increase to continue to the highest value. It showed that for women, risk was flat at 3.65 percent for a glucose level of 5.55 mmol/liter, then it increased rapidly through the impaired range (reaching 8.34 percent at 6.99 mmol/liter) and increased much more slowly thereafter (figure 3). In the women-spline model, the middle slope was significant (p = 0.008) and the difference between the middle and right slopes was significant (p = 0.033), with the middle slope being 9.34 times as large as the right slope. The right slope was positive and nearly significant (p = 0.112). The unrestricted spline with these knots supported the lack of association of risk with glucose in the normal range. In that model, the left slope was very small, negative, and nonsignificant (p = 0.857). Rates predicted above the diabetes cutpoint should be considered cautiously because of a lack of data.

The interval of constancy in the mortality/glucose relation for men cannot be attributed to lack of data. Indeed, for men, the numbers of observations and deaths in the above-normal range were greater than those for women (table 2). Similarly, in the normal range, there were only 38 percent more observations for men than for women, which is insufficient to account for the dramatic difference between men and women observed in this range.

Cardiovascular and noncardiovascular mortality
We separately examined cardiovascular and noncardiovascular mortality by using the same methods as for all-cause mortality. As would be expected from the fact that 69.7 percent of the deaths were cardiovascular related, exactly analogous relations held for cardiovascular mortality as for all-cause mortality (data not shown). For non-cardiovascular-related death, data were insufficient to reliably examine the relation to glucose for men and women separately, to discriminate various models for men and women combined, or to examine risk in our nine glucose categories. Interestingly, there was evidence that noncardiovascular mortality may also be related to glucose. Table 2 gives the number of non-cardiovascular-related deaths in the normal, impaired, and diabetic categories. The adjusted non-cardiovascular-related death rates in these categories steadily increased from 1.78 percent (95 percent confidence interval: 1.28 percent, 2.49 percent) to 2.66 percent (95 percent confidence interval: 1.79 percent, 3.95 percent) to 2.96 percent (95 percent confidence interval: 1.77 percent, 4.91 percent), respectively, but the overall difference was not significant (p = 0.148). Examining for a linear trend, which has more power than the previous procedure to detect an increasing association, showed that, overall, there was a significant upward linear trend of noncardiovascular mortality to glucose (p = 0.009) in which risk increased from 1.82 percent at 3.89 mmol/liter to 2.06 percent at 5.55 mmol/liter to 2.29 percent at 6.99 mmol/liter, reaching 3.07 percent at 11.1 mmol/liter. When analysis was restricted to the nondiabetic range, for which we had very limited data, the significance of the trend deteriorated (p = 0.097).

Additional analyses
Adding interactions between glucose and other covariates to the models showed that no interactions were significant. None of our results on the mortality/glucose relations in the subdiabetic range materially changed when glucose-intolerant (and therefore diabetic) subjects were omitted from the analyses, and none of our findings were altered when antihypertensive drug users were omitted. The large difference in smoking between men and women (table 1) did not account for the large differences we observed between the sexes. For either sex, smoking was not significantly correlated with glucose, and the same differences between men and women persisted when the analyses were restricted to nonsmokers.

Fasting versus random glucose
Potentially, use of random instead of fasting glucose could have introduced additional random error due to variability in time since the last meal that would compromise our analyses by diluting the associations of risk with glucose. Clearly, this problem did not occur because we found very strong associations. Additionally, blood was drawn at the conclusion of many hours of examination, and one would expect that the random and fasting values would be close. Fortunately, we can verify that this was the case. Fasting glucose values were available at four of the examinations. The age- and sex-adjusted deciles of the fasting and random glucose distributions were very close (indicating that the two distributions were close), with the random ones being slightly lower. Additionally, t tests for the difference between random and fasting glucose for each age-sex group showed no significant differences. Thus, in our case, the use of random instead of fasting glucose had little impact on our analyses.


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
APPENDIX
 References
 
The Framingham data enabled us to examine the mortality/glucose relation in a population that had not yet been aggressively treated with agents known to affect CVD (e.g., statins, angiotensin-converting enzyme inhibitors) or with oral antihyperglycemic agents that might confound that relation. While our results show that glucose is a strong, independent predictor of mortality, they also unequivocally show that the mortality/glucose relations are qualitatively different for men and women. Indeed, for men, risk rises very rapidly through the normal range and is flat thereafter; for women, risk is unrelated to glucose in the normal range, then rises quickly through the impaired range and at a much slower rate thereafter (figure 4). To our knowledge, such striking differences between the sexes have not been previously reported in subjects with CVD.

Prior investigations of the mortality/glucose association in subjects with chronic CVD examined only men and women combined (4Go, 5Go, 7Go). Although our findings for men and women combined are in accord with the findings from these studies, they also clearly indicate that the mortality/glucose relation for men and women combined gives a biased view of the relations for either sex individually. When both sexes were considered together, the rapid rise seen through the normal range was due to the men, while the rise in the above-normal range was due to the women (figures 2 and 4).

Although speculations have been advanced on how glucose affects mortality, this mechanism remains basically unknown. Glucose may have direct harmful effects on vascular endothelium or atherosclerotic plaque, mediated by nonenzymatic glycosylation of apolipoproteins and clotting factors, increased oxidative stress, and activation of the polyol pathway (14Go). Alternatively, insulin may be the factor that worsens CVD, with elevated glucose levels driving compensatory hyperinsulinemia (15Go). The dramatic increases in mortality risk observed with glucose levels in the subdiabetic range are in accord with the concept that much cardiovascular risk accrues before overt diabetes develops, because the prediabetic state is characterized by inflammation and endothelial dysfunction (16Go, 17Go). Why the relations should be so different for the two sexes is unknown and needs further study.

We caution that our spline models attempt to capture smooth curvilinear relations of mortality to glucose, albeit somewhat artificially, by piecewise-linear approximations and should not be interpreted too literally. In particular, the knots, and the abrupt changes at them, are artifacts of the model, and the knots do not necessarily represent "thresholds" of any sort. Despite their shortcomings, the adjusted rates predicted by the spline models in each of the normal, impaired, and diabetic categories, as well as the nine categories, were in good agreement with the adjusted observed rates in these categories (figure 1 and table 3).


View this table:
[in this window]
[in a new window]
  		TABLE 3. Adjusted 2-year mortality rates/100, by glucose level, predicted by the nine-categories and spline models,* Framingham Heart Study, 1948–1978

 
There are some limitations to our investigation. Our findings may not apply to subjects in racial and socioeconomic groups different from those in the Framingham population (mostly White, middle class) or to subjects younger or older (aged <45 or >74 years, respectively) than those in our population. We were unable to assess the roles of insulin, high density lipoprotein cholesterol, low density lipoprotein cholesterol, triglycerides, and glycosylated hemoglobin (HbA1c) because these factors were not measured in the Framingham Heart Study. Data were insufficient to examine the mortality/glucose relation separately for various types of CVD.

Our findings might partially explain why clinical trials of glucose lowering in diabetic subjects, such as the Diabetes Control and Complications Trial (18Go), the United Kingdom Prospective Diabetes Study (19Go), and others, have not definitively demonstrated mortality reduction with glucose lowering; in these trials, the intensive control groups achieved mean HbA1c values of only about 7 percent, which corresponds to a glucose level of 9.6 mmol/liter (172 mg/dl) (20Go). According to our results, such reductions would have no effect for men and only a very small effect for women.

Strictly speaking, our findings apply to only those diabetic subjects with recognized CVD; however, it is reasonable to assume that they apply to all diabetics because most probably have substantial (but as yet unrecognized) CVD (21Go). Thus, our findings underscore the need for aggressive glucose-lowering trials that aim at reductions well into the normal range, especially for men. Only such trials can determine whether the strong mortality/glucose associations reported here are causal.

In conclusion, our study revealed possible, strong gender-based differences in the association between blood glucose levels and mortality. Findings indicate that the diabetic cutpoint (6.99 mmol/liter) does not provide a useful mortality risk division, especially for men, although the normal cutpoint (5.55 mmol/liter) may be a useful division, albeit with quite different implications, for men than women. These findings suggest that future studies are needed to examine gender differences with regard to the prediabetic state, the development of diabetes, and CVD.


   APPENDIX
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
References
 
The general form for a spline model was f(g) = {alpha} + {gamma}1g + {gamma}2 max(0,g – k1) + {gamma}3 max(0,g – k2) + {gamma}4 max(0,g – k3), where one or more of the {gamma}1, ..., {gamma}4 can be 0 and constraints may be placed to force the right-most slope (R) or left-most slope (L) to be 0. We specify a particular spline as spl(knot1,knot2, ...; restrictions). For example, spl(5.55,6.99; R = 0) is the spline with knots at 5.55 and 6.99 mmol/liter and the restriction that the right-most slope = 0; spl(5.55,6.99) is the spline with these knots and no slope restrictions.

AIC is defined as –2ll + 2p, where ll is the log likelihood of the model and p the total number of parameters (including intercept). Judged by AIC, the optimal model is the one with the smallest AIC. Models within 1 of the optimal model were considered equivalent to it. Appendix table 1 gives the formulas and AIC of the models we considered.
APPENDIX TABLE 1. Log likelihood and AIC* of models considered

Model
Men and women
Men
Women
ll(model)*
 df
 AIC
 ll(model)
 df
 AIC
 ll(model)
 df
 AIC
Linear [f(g) = {alpha} + {gamma}g] –886.318 13 1,798.64 –593.38 10 1,206.77 –287.08 10 594.16
Linear with sex-glucose interaction –884.605 14 1,797.21
Equilibrium [f(g) = {alpha} + {gamma}(g/1 + g)] –882.251 13 1,790.50 –590.41 10 1,200.82 –286.71 10 593.42
Log [f(g) = {alpha} + {gamma}ln(g)] –883.775 13 1,793.55 –591.81 10 1,203.63 –286.42 10 592.85
Sqrt Formula –885.044 13 1,796.09 –592.69 10 1,205.38 –286.61 10 593.21
Quad [f(g) = {alpha} + {gamma}1g + {gamma}2g2] –884.796 14 1,797.59 –592.18 11 1,206.36 –286.55 11 595.10
Lin_log [f(g) = {alpha} + {gamma}1g + {gamma}2 ln(g)] –882.469 14 1,792.94 –589.48 11 1,200.97 –286.42 11 594.84
Lin_sqrt Formula –883.071 14 1,794.14 –590.23 11 1,202.47 –286.42 11 594.84
Spl(5.55) –880.816 14 1,789.63{dagger},{ddagger} –585.82 11 1,193.64 –287.00 11 596.01
Spl(6.11) –880.708 14 1,789.42{dagger} –587.07 11 1,196.14 –286.63 11 595.26
Spl(6.99) –881.303 14 1,790.61 –588.68 11 1,199.36 –286.14 11 594.28
Spl(5.55,6.11) –880.621 15 1,791.24 –585.69 12 1,195.39 –285.22 12 594.45
Spl(5.55,6.99) –880.581 15 1,791.16 –585.73 12 1,195.46 –285.17 12 594.35
Spl(6.11,6.99) –880.708 15 1,791.42 –586.65 12 1,197.30 –285.60 12 595.20
Spl((5.55,6.11,6.99) –880.579 16 1,793.16 –585.69 13 1,197.39 –285.07 13 596.13
Spl(5.55; R = 0) –882.554 13 1,791.11 –585.83 10 1,191.66{dagger},{ddagger} –292.16 10 604.32
Spl(6.11; R = 0) –881.607 13 1,789.21{dagger} –587.16 10 1,194.31 –290.35 10 600.70
Spl(6.99; R = 0) –881.576 13 1,789.15{dagger} –588.91 10 1,197.81 –288.19 10 596.38
Spl(5.55,6.11; R = 0) –881.607 14 1,791.21 –585.70 11 1,193.40 –287.50 11 596.99
Spl(5.55,6.11; R = 0) –881.164 14 1,790.33 –585.74 11 1,193.49 –286.28 11 594.57
Spl(6.11,6.99; R = 0) –881.319 14 1,790.64 –586.67 11 1,195.34 –286.69 11 595.39
Spl(5.55; L = 0) –888.099 13 1,802.20 –594.35 10 1,208.70 –287.36 10 594.73
Spl(6.11; L = 0) –888.711 13 1,803.42 –594.52 10 1,209.05 –287.84 10 595.67
Spl(6.99; L = 0) –889.323 13 1,804.65 –594.65 10 1209.30 –288.47 10 596.94
Spl(5.55,6.11; L = 0) –884.338 14 1,796.68 –592.33 11 1,206.66 –285.31 11 592.61{dagger}
Spl(5.55,6.99; L = 0) –885.066 14 1,798.13 –593.00 11 1207.99 –285.19 11 592.38{dagger},{ddagger}
SSpl(6.11,6.99; L = 0)
 –886.45
 14
 1,800.90
 –593.75
 11
 1,209.51
 –285.75
 11
 593.49{dagger}

* AIC, Akaike Information Criterion; ll(model), log-likelihood of the model.

{dagger} Model with optimal AIC.

{ddagger} The model selected.


   ACKNOWLEDGMENTS
 
The authors are indebted to the National Institutes of Health (Bethesda, Maryland) for providing the Framingham data.

Glaxo-Smith Kline Pharmaceutical (Clifton, New Jersey), Merck Pharmaceutical (Rahway, New Jersey), Takeda Pharmaceutical (Deerfield, Illinois), Novartis Pharmaceutical (East Hanover, New Jersey), and Boehringer Ingelheim Pharmaceutical (Ridgefield, Connecticut) provided research grant support to Dr. Hsueh.


   References
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 APPENDIX
 References
 

  1. Capes SE, Hunt D, Malmberg K, et al. Stress hyperglycaemia and increased risk of death after myocardial infarction in patients with and without diabetes: a systematic overview. Lancet 2000;355:773–8.[CrossRef][Web of Science][Medline]
  2. Zindrou D, Taylor KM, Bagger JP. Admission plasma glucose: an independent risk factor in nondiabetic women after coronary artery bypass grafting. Diabetes Care 2001;24:1634–9.[Abstract/Free Full Text]
  3. Zeller M, Cottin Y, Brindisi MC, et al. Impaired fasting glucose and cardiogenic shock in patients with acute myocardial infarction. Eur Heart J 2004;25:308–12.[Abstract/Free Full Text]
  4. Fisman EZ, Motro M, Tenenbaum A, et al. Impaired fasting glucose concentrations in nondiabetic patients with ischemic heart disease: a marker for a worse prognosis. Am Heart J 2001;141:485–90.[CrossRef][Web of Science][Medline]
  5. Muhlestein JB, Anderson JL, Horne BD, et al. Effect of fasting glucose levels on mortality rate in patients with and without diabetes mellitus and coronary artery disease undergoing percutaneous coronary intervention. Am Heart J 2003;146:351–8.[CrossRef][Web of Science][Medline]
  6. Arcavi L, Behar S, Caspi A, et al. High fasting glucose levels as a predictor of worse clinical outcome in patients with coronary artery disease: results from the Bezafibrate Infarction Prevention (BIP) study. Am Heart J 2004;147:239–45.[CrossRef][Medline]
  7. Port SC, Goodarzi MO, Boyle NG, et al. Blood glucose: a strong risk factor for mortality in non-diabetic patients with cardiovascular disease. Am Heart J 2005;150:209–14.[CrossRef][Medline]
  8. Dawber TR. The Framingham Study: the epidemiology of atherosclerotic disease. Cambridge, MA: Harvard University Press, 1980.
  9. Nelson NA. Photometric adaptation of the Somogyi method for the determination of glucose. J Biol Chem 1944;153:375–80.[Free Full Text]
  10. Kuwa K, Nakayama T, Hoshino T, et al. Relationships of glucose concentrations in capillary whole blood, venous whole blood and venous plasma. Clin Chim Acta 2001;307:187–92.[CrossRef][Web of Science][Medline]
  11. D'Agostino RB, Lee ML, Belanger AJ, et al. Relation of pooled logistic regression to time dependent Cox regression analysis: the Framingham Heart Study. Stat Med 1990;9:1501–15.[Web of Science][Medline]
  12. Cupples LA, D'Agostino RB, Anderson K, et al. Comparison of baseline and repeated measure covariate techniques in the Framingham Heart Study. Stat Med 1988;7:205–22.[Web of Science][Medline]
  13. Burnham KP, Anderson DR. Model selection and multimodel inference. New York, NY: Springer-Verlag, 2002.
  14. King GL, Wakasaki H. Theoretical mechanisms by which hyperglycemia and insulin resistance could cause cardiovascular diseases in diabetes. Diabetes Care 1999;22(suppl 3):C31–7.
  15. Pyorala M, Miettinen H, Laakso M, et al. Hyperinsulinemia predicts coronary heart disease risk in healthy middle-aged men: the 22-year follow-up results of the Helsinki Policemen Study. Circulation 1998;98:398–404.[Abstract/Free Full Text]
  16. Hsueh WA, Quinones MJ. Role of endothelial dysfunction in insulin resistance. Am J Cardiol 2003;92:10J–17J.[Web of Science][Medline]
  17. Haffner SM. Insulin resistance, inflammation, and the prediabetic state. Am J Cardiol 2003;92:18J–26J.[Web of Science][Medline]
  18. The effect of intensive treatment of diabetes on the development and progression of long-term complications in insulin-dependent diabetes mellitus. The Diabetes Control and Complications Trial Research Group. N Engl J Med 1993;329:977–86.[Abstract/Free Full Text]
  19. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33). UK Prospective Diabetes Study (UKPDS) Group. Lancet 1998;352:837–53.[CrossRef][Web of Science][Medline]
  20. Rohlfing CL, Wiedmeyer HM, Little RR, et al. Defining the relationship between plasma glucose and HbA(1c): analysis of glucose profiles and HbA(1c) in the Diabetes Control and Complications Trial. Diabetes Care 2002;25:275–8.[Abstract/Free Full Text]
  21. Haffner SM, Lehto S, Ronnemaa T, et al. Mortality from coronary heart disease in subjects with type 2 diabetes and in nondiabetic subjects with and without prior myocardial infarction. N Engl J Med 1998;339:229–34.[Abstract/Free Full Text]

Add to CiteULike CiteULike   Add to Connotea Connotea   Add to Del.icio.us Del.icio.us    What's this?


This article has been cited by other articles:


Home page
 JAMAHome page
M. O. Goodarzi and B. M. Psaty
Glucose Lowering to Control Macrovascular Disease in Type 2 Diabetes: Treating the Wrong Surrogate End Point?
JAMA, November 5, 2008; 300(17): 2051 - 2053.
[Full Text] [PDF]

Home page
 Journals of Gerontology Series A: Biological Sciences and Medical SciencesHome page
A. M. Kulminski, S. V. Ukraintseva, I. V. Culminskaya, K. G. Arbeev, K. C. Land, L. Akushevich, and A. I. Yashin
Cumulative Deficits and Physiological Indices as Predictors of Mortality and Long Life
J. Gerontol. A Biol. Sci. Med. Sci., October 1, 2008; 63(10): 1053 - 1059.
[Abstract] [Full Text] [PDF]

This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow All Versions of this Article:
163/4/342    most recent
kwj027v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (4)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Port, S. C.
Right arrow Articles by Goodarzi, M. O.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Port, S. C.
Right arrow Articles by Goodarzi, M. O.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?