| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
RESEARCH AND PRACTICE |
The author is with the Congressional Budget Office, Washington, DC.
Correspondence: Requests for reprints should be sent to Jerome Timothy Gronniger, MPP, MHSA, Ford House Office Building, 2nd and D Sts, SW, Washington, DC 20515 (e-mail: tim. gronniger{at}cbo.gov).
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
Objectives. I used a semi-parametric analysis of the relationship between body mass index (BMI) and mortality to assess the adequacy of conventional BMI categories for planning public health programs to reduce mortality.
Methods. I linked supplements from the 1987 and 1989 versions of the National Health Interview Survey to the 1995 Multiple Cause of Death File to obtain mortality information. I constructed nonlinear estimates of the association between BMI and mortality using a semiparametric regression technique.
Results. The mortality risk among "normal" weight men (i.e., those in the BMI range of 20 to 25 kg/m2) was as high as that among men in the mild obesity category (BMIs of 30–35 kg/m2), with a minimum risk observed at a BMI of approximately 26 kg/m2. Among women, the mortality risk was smallest at approximately 23 to 24 kg/m2, with the risk increasing steadily with BMIs above 27 kg/m2. In each specification, the slope of the line was small and volatile through the BMI range of 20 to 35 kg/m2, suggesting negligible risk differences with minor differences in weight for much of the population.
Conclusions. Traditional BMI categories do not conform well to the complexities of the BMI–mortality relationship. In concurrence with conclusions from previous literature, I found that the current definitions of obesity and overweight are imprecise predictors of mortality risk.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
Obesity is but 1 of the possible categories along the BMI continuum; studies have compared differences in mortality and morbidity risks18 between individuals who are mildly, severely, and morbidly obese, with typical respective demarcations at BMIs of 30, 35, and 40 (corresponding to class I, class II, and class III obesity under alternate terminology19). When BMI categories are segmented in this fashion, most of the literature is in agreement that obesity is associated with clear increases in risk of mortality, while overweight is a risk factor for obesity and thus best avoided.2,4,15
However, relying on broad categories such as overweight and obesity could provide misleading estimates of BMI’s association with mortality if that association is heterogeneous or not monotonic within categories. As is the case with all binary risk factors based on continuous measures, obesity is an arbitrary category by necessity. Its definition grew out of a consensus among various health bodies (including the World Health Organization, the National Institutes of Health, and the Centers for Disease Control and Prevention) that health risks increase with increasing body weight above a BMI of 25 and become serious near a BMI of 30.17 Individuals categorized in these groups have been termed overweight and obese, respectively. Optimal BMIs are believed to be in the range of 20 to 25, and thus this range is defined as normal. The definition of underweight is fluid but is usually set at 18.5.14
Weight differentials between the BMI categories can be quite large. If obese individuals are seeking to attain the "optimal" BMI, a weight loss of more than 30 lb (13.5 kg) might be necessary. Conversely, individuals with a BMI of 29 who gain 5 to 9 lb (2.2 to 4 kg) might find themselves newly classified as obese and at high risk for early mortality and various diseases. These categories have generally yielded results such as those illustrated in Figure 1
, which shows a "U"- or "J"-shaped relationship between BMI and mortality. The consistency of these estimates is not as well established within the categories themselves, nor is whether the choice of category width ("bin width") unduly influences study findings.
|
and Figure 2. In 2002 alone, there were 17 million adults in the United States with BMIs between 30 and 32. Thus, it is incumbent on health officials to carefully consider the impact of their guidance on the public. Some studies have parsed out the BMI–mortality relationship by examining categories narrower than 5 BMI units,11 but none have provided a picture of the full BMI continuum.16
|
|
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
I treated BMI nonparametrically and entered all other covariates into the model in a parametric fashion. Formally, mortality was modeled as dependent on a parametric term, Ziß, and a nonparametric function of body mass index, m(BMIi):
 | (1) |
Deathi, where i indexes individuals, is an indicator variable describing whether an individual died before the end of follow-up (i.e., 1995 in this study). Zi is a matrix of co-variates including age categories, gender, race, education level, income level, survey processing year, and current smoking status, and ß is its associated parameter vector. These variables were selected because they are frequently used in the medical literature on obesity (age, race, smoking) or because they were highly correlated with obesity in a cross-sectional analysis (education level and income level). m(BMIi) is the independent nonparametric effect of BMI on mortality.20
Initially, the "effect" of BMI is parsed out by sorting all variables by BMI and taking the difference between each observation and the observation preceding it (i.e., Zi – Zi–1), a process termed "taking first differences." Deathi is then regressed on the differenced Zi to estimate 
. The parameter vector is used to generate predicted values of Deathi with Zi. Subtracting the predicted values from the actual values (Deathi Zi – ) generates a dependent variable "purged" of the effects of Zi, and this variable is then used in a locally weighted regression (conducted via the Lowess procedure in Stata [Stata Corp, College Station, Tex]) on BMI to estimate m (BMIi).
A small bandwidth of 0.15 was selected to capture local variations in the data. In analyses involving smaller bandwidths, a more local slice of data is used to estimate m(BMIi), and therefore a less smooth result is obtained. Estimating the average parametric term (Zi – ) and adding the nonparametric effect of BMI—m(BMIi)—for each individual produces an estimate of BMI-specific expected mortality for an individual with variable BMIi but average values for the covariates age, race, gender, education, income, current smoking status, and processing year (Zi – ).
It is important to note that the intercept in a nonparametric regression has no natural interpretation; only relative differences are estimated. I used an "average" individual (along all covariates in Zi) to center the distribution on a reasonable expected mortality level, which aids interpretation but does not make point estimates of mortality meaningful. This limitation is not severe: relative differences determine local minima and maxima, the putative "optimal BMI" levels in regard to mortality. Similarly, it is not appropriate to compare mortality point estimates between specifications. It is possible, however, to compare marginal differences and thus examine the slope of the line in various regions of the BMI distribution.
Data Sources
The 1987 Cancer Control and 1989 Diabetes supplements of the National Health Interview Survey (NHIS) were used to obtain baseline personal and biometric information on 1987 and 1989 survey respondents.23,24 The NHIS involves in-person interviews conducted each year by the National Center for Health Statistics to track the health characteristics of the civilian, noninstitutionalized population. The Cancer Control and Diabetes supplements are restricted to the adult population 18 years or older. It was necessary to use these 2 supplements rather than core files because the NHIS did not include tobacco use questions until the 1990s. The pooled data from the 2 surveys provided 106 615 observations for which only limited smoking information was available. These files were linked to the NHIS Multiple Cause of Death File to obtain mortality follow-up information through 1995.25
After the sample had been restricted to individuals with data on BMI and smoking status, and those older than 64 years at baseline had been eliminated, baseline measurements and mortality follow-up information were available for 33 558 individuals, of whom 1109 were dead or presumed dead. (Sensitivity analyses not described here involved a larger sample of 101 830 respondents for whom smoking data were not available. This sample included 7327 deaths occurring before or during 1995.) Comparison of the final sample with the original sample revealed few statistically significant differences in regard to the covariates assessed, although the final (included) sample contained relatively fewer women, and, overall, education levels were lower. (Descriptive statistics on included and excluded respondents are available upon request from the author.) For these reasons, the final sample was not completely representative of the US adult civilian population younger than 65 years in 1987 and 1989, but it did not deviate sharply from the full NHIS sample, which was representative of the entire population.
Use of sample weights adjusted somewhat for the oversampling techniques employed by the National Center for Health Statistics. However, weights can be applied to estimates of Zi ß (Equation 1) but not to m(BMI)i, so the final semiparametric estimates were not fully weighted and the results cannot be deemed nationally representative. Because no hypotheses were tested, it was not necessary to correct standard errors for survey design effects.
Individuals with incomplete mortality or BMI data were excluded from the analyses; all other missing values were coded as separate categories and did not result in exclusions. Complete smoking data were not available, but respondents were coded as "current" or "not current" smokers.
Prevalence estimates were obtained from the 2002 NHIS Sample Adult File.26 This file contains information on 31 044 respondents, among whom 29 417 have BMI data available. A kernel density estimator was used to calculate population densities in single-unit BMI intervals (i.e., 21–22, 22–23, and so forth). Estimated densities were applied to the sums of the sample weights for 2002 to obtain estimated population frequencies for these cells.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
plots the estimated relationships between mortality and BMI among men, women, and all NHIS respondents who were aged between 18 and 64 years in 1987 or 1989. These relationships were complex and did not fall into the conventional BMI categories. Contrary to predictions of high mortality risks among overweight individuals, the "optimum" BMI appears to be 26 to 27 overall, 23 to 24 for women, and 26 to 27 for men. Interestingly, in the case of all 3 groups, mortality did not increase sharply with BMI until the range of about 27 or above (33–35 for men), well into the ambit of overweight and obesity.
|
suggest that mortality among clinically underweight individuals is quite high, although estimates near the tails of the BMI distribution are imprecise as a result of small local sample sizes. Men in the "normal" weight category exhibited a mortality rate as high as that of men in the moderately obese category (BMIs of 30 to 35), and men in the "overweight" category clearly had the lowest mortality risk, although a sharp jump was observed between BMIs of 28 and 30. Among women, the predicted shape held better, but this was not entirely the case. Expected mortality steadily increased from a BMI of 27 through the range of obesity, but even at a BMI of 32 it was possible to locate a point of equal predicted mortality in the normal region: at a BMI of 20. Normal weight individuals of both genders did not appear to be relatively more long-lived than mildly obese individuals (BMIs of 30–35), whereas overweight people (BMIs of 25–30) appeared healthiest of all.
Perhaps more remarkable than any particular trend in the data are the flatness and instability of the curve over the most common BMI ranges. The flattest regions, between BMIs of 20 and 35, included the most people. In this pooled sample from 1987 and 1989, 85.7% of the population fell in the 20 to 35 BMI range, and 96.6% of these individuals had BMIs below 35. The numbers were similar for more recent time periods (85.9% had BMIs between 20 and 35 in 2002, while 90.6% had BMIs below 35) (Table 1). Thus, a large fraction of the population clusters near the average BMI of 25. Because the BMI–mortality slope is gentle in areas where population density is thickest, this work does not support the idea that reducing weight alone would result in any large mortality risk reduction for most of the population.
There were limitations involved with this study, including the arbitrary character of the intercept estimates, which were derived through a semiparametric approach (as described in the "Methods" section). This complicates interpretation but did not bias the results. As has been the case with other studies focusing on the relationship between BMI and mortality, the present study is unable to solve the heterogeneity problem; even after age, gender, education level, and other covariates had been controlled for, BMI might have been tied to manifold variables that influence mortality. Many of these omitted risk factors might be correlated with BMI, leading to misestimation of the risk of increasing BMI itself. For this reason, no study has actually identified the mortality-minimizing or "optimal" BMI.
Furthermore, optimal BMI might be expected to be contingent on current weight; for example, there is no guarantee that losing weight will bring a severely obese person’s expected mortality down to the optimal level. Studies involving single-point-in-time measures of BMI cannot unravel this problem. However, virtually every study on the matter has worked from the same research design employed here, namely, a prospective, nonexperimental design with a single-point-in-time measurement. Studies forming the basis for recommendations of weight loss to obese and overweight individuals have relied on even stronger assumptions than this work.5,8–11,27
Because of the absence of standard errors, the semiparametric estimates presented here cannot be used in hypothesis testing; thus, for example, the expected mortality at a BMI of 29.99 cannot be statistically compared with the expected mortality at a BMI of 30.01. In addition, while the NHIS is nationally representative and can provide externally valid point estimates when sample weights are used, the construction of the data set used here and the limits of the semiparametric method prevent full weighting of the mortality-by-BMI point estimates. Therefore, the present results are not truly nationally representative, although comparisons of the included and excluded groups suggest that the sample largely resembles the US adult civilian, noninstitutionalized population younger than 65 years at the time of this study.
A final but important limitation relates to the striking peaks and valleys that can be observed in Figure 3, which seem to invite close inspection and interpretation of each bit of nonlinearity. Such parsing probably approaches overinterpretation. The lumpiness of the BMI–mortality plot represents noisy data as well as true associations, so inferences regarding the effects of infinitesimal changes in BMI are precarious. The present results can better be used to consider broad trends over at least several BMI units and to contrast such trends with findings derived from categorical studies.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
The present results highlight previous findings indicating that mild obesity and overweight are not strongly related to mortality.16,28 In addition, one cannot assume that a risk measured for a person with a BMI of 35 should concern a person with a BMI of 30, especially when the personal and economic costs of weight loss are high. It would be more reasonable to focus on the smaller group of people in the severely obese category (BMIs of 40 and over; 3.3% of adults in 2002), which has a clearer relationship to mortality; however, some of this association, too, might be contaminated by omitted variables.
Beyond problems of categorization, the apparently small differences in mortality among overweight and obese individuals with small differences in BMI again call into question the effectiveness of weight loss strategies in regard to decreasing mortality. Guidelines issued by the National Heart, Lung, and Blood Institute and other medical bodies call for overweight individuals to aim for weight loss regardless of their current weight.29–31 The target is thus the same for men who are 5 lb overweight and women who are 100 lb overweight: to lose 10% of their body weight, maintain the loss, and reevaluate their weight,31 presumably with the goal of beginning another weight loss program if a "normal" weight has not been achieved. However, these guidelines disregard evidence indicating not only that weight loss is rare and difficult even in controlled diet settings32–34 but that it might also be harmful.
Several authors have reported that weight loss is actually associated with increased mortality.35–37 In an interesting corollary, Gregg et al. reported that attempting to lose weight is beneficial for health—whether or not weight is actually lost.38 Such results beg the question of when and how weight loss is relevant to one’s health. Current evidence strongly supports the beneficial effects of diet and exercise33,39,40 but does not provide convincing proof of the benefits of losing weight. The results of the studies just described are consistent with the flat gradient of mortality with BMI documented here.
The subtleties of this relationship are not at all apparent from analyses involving large BMI categories. The semiparametric approach used here provides a clearer picture of individual mortality risks because restrictive categories were eliminated and the data were allowed to shape the functional form. The distinctions between overweight, obesity, and severe obesity have frequently been neglected, with overweight often assumed to be dangerous simply because it approaches obesity. Attention to such fine distinctions is important, because each BMI unit in the 25 to 30 range involves literally tens of millions of people. It would be as great a mistake to overdraw the overweight and obese categories as to underclassify them, and thus it seems best to avoid exaggerating the mortality risks faced by individuals with BMIs below 35.
In summary, the semiparametric approach used in this study provides a lens on the BMI–mortality problem that complements previous work with BMI categories. The present results are consistent with earlier findings documenting very small to negligible increased mortality risks among overweight and mildly obese individuals, especially in the case of men. The lower reaches of the overweight and obesity spectrum seem far less dangerous, in terms of mortality at least, than is generally advertised. In addition, the target BMI category of 20 to 25 sometimes carries a higher risk of mortality than portions of the 25 to 30 and 30 to 35 categories. These results raise questions about whether overweight and mildly obese individuals are classified correctly under current health guidelines. Health professionals should bear in mind both the large number of people involved and the modest mortality differences between BMI units in drafting health guidelines and planning public health programs.
|
Acknowledgments |
|---|
I thank John DiNardo for essential theoretical guidance and technical advice. Many of this report’s ideas originated with him. I also thank Catherine McLaughlin for assistance in revising the article and Eric Klein for editing assistance. Mike Chernew’s comments on other versions of the article aided the conceptual approach.
Human Participant Protection
No protocol approval was needed for this study.
|
Footnotes |
|---|
Note. The views expressed in this article are those of the author and should not be interpreted as those of the Congressional Budget Office. In addition, neither the National Center for Health Statistics nor the Inter-University Consortium for Political and Social Research bears any responsibility for the analysis or interpretations presented here.
Accepted for publication July 14, 2004.
|
References |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONReferences |
|---|
2. Allison DB, Fontaine KR, Manson JE, Stevens J, VanItallie TB. Annual deaths attributable to obesity in the United States. JAMA. 1999;282:1530–1538.
3. Allison DB, Znnolli R, Venkat-Narayan KM. The direct health care costs of obesity in the United States. Am J Public Health. 1999;89:1194–1199.
4. National Task Force on the Prevention and Treatment of Obesity. Overweight, obesity, and health risk. Arch Intern Med. 2000;160:898–904.
5. Dorn JM, Schisterman EF, Winkelstein W, Trevisan M. Body mass index and mortality in a general population sample of men and women: the Buffalo Health Study. Am J Epidemiol. 1997;146:919–931.
6. Yuan JM, Ross RK, Gao YT, Yu MC. Body weight and mortality: a prospective evaluation in a cohort of middle-aged men in Shanghai, China. Int J Epidemiol. 1998;27:824–832.
7. Lee IM, Manson JE, Hennekens CH, Paffenbarger RS. Body weight and mortality: a 27-year follow up of middle age men. JAMA. 1993;270:2823–2828.[Abstract]
8. Manson JE, Willett WC, Stampfer MJ, et al. Body weight and mortality among women. N Engl J Med. 1995;333:677–685.
9. Fontaine KR, Redden DT, Wang C, Westfall AO, Allison DB. Years of life lost because of obesity. JAMA. 2003;289:187–193.
10. Harris T, Cook EF, Garrison R, Higgins M, Kannel W, Goldman L. Body mass index and mortality among nonsmoking older persons: the Framingham Heart Study. JAMA. 1998;259:1520–1524.
11. Calle EE, Rodriguez C, Walker-Thurmond K, Thun MJ. Overweight, obesity, and mortality from cancer in a prospectively studied cohort of US. adults. N Engl J Med. 2003;348:1625–1638.
12. Sturm R. The effects of obesity, smoking, and drinking on medical problems and costs. Health Aff. 1993;12:245–253.[CrossRef]
13. The Surgeon General’s Call to Action to Prevent and Decrease Overweight and Obesity. Washington, DC: US Dept of Health and Human Services; 2001.
14. Flegal KM, Troiano RP, Pamuk ER, Kuczmarski RJ, Campbell SM. Prevalence and trends in obesity among US adults, 1999–2000. JAMA. 2002;288: 1723–1727.
15. Calle EE, Thun MJ, Petrelli JM, Rodriguez C, Heath CW Jr. Body mass index and mortality in a prospective cohort of US. adults. N Engl J Med. 1999;341: 1097–1105.
16. Durazo-Arvizu R, McGee D, Zhaohai L, Cooper R. Establishing the nadir of the body mass index-mortality relationship: a case study. J Am Stat Assoc. 1997;92: 1312–1319.[CrossRef][ISI]
17. Kuczmarksi RJ, Flegal KM. Criteria for definition of overweight in transition: background and recommendations for the United States. Am J Clin Nutr. 2000;72:1074–1081.
18. Must A, Spadano J, Coakley EH, Field AE, Colditz G, Dietz WH. The disease burden associated with overweight and obesity. JAMA. 1999;282:1523–1529.
19. Freedman DS, Khan LK, Galuska DA, Dietz WH. Trends and correlates of class 3 obesity in the United States from 1990 through 2000. JAMA. 2002;288: 1758–1761.
20. DiNardo J, Tobias JL. Nonparametric density and regression estimation. J Econ Perspect. 2001;15:11–28.
21. Yatchew A. An elementary estimator of the partial linear model. Econ Lett. 1997;57:135–143.[CrossRef][ISI]
22. Yatchew A. Non-parametric regression techniques in economics. J Econ Lit. 1998;36:669–721.
23. National Center for Health Statistics. National Health Interview Survey, 1987: Cancer Control Study. Ann Arbor, Mich: Inter-University Consortium for Social and Political Research; 1989.
24. National Center for Health Statistics. National Health Interview Survey, 1989: Diabetes Supplement. Ann Arbor, Mich: Inter-University Consortium for Social and Political Research; 1991.
25. National Center for Health Statistics. National Health Interview Surveys, 1986–1994: Multiple Cause of Death, Dates of Death 1986–1995. Ann Arbor, Mich: Inter-University Consortium for Social and Political Research; 1997.
26. Data File Documentation, National Health Interview Survey, 2002. Hyattsville, Md: National Center for Health Statistics; 2003.
27. Rogers RG, Hummer RA, Krueger PM. The effect of obesity on overall circulatory disease and diabetes-specific mortality. J Biosoc Sci. 2003;35:107–129.[CrossRef][ISI][Medline]
28. Lantz PM, House JS, Lepkowski JM, Williams DR, Mero RP, Chen J. Socioeconomic factors, health behaviors, and mortality. JAMA. 1998;279:1703–1708.
29. Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults: The Evidence Report. Bethesda, Md: National Heart, Lung, and Blood Institute; 1998.
30. National Heart, Lung, and Blood Institute. Aim for a healthy weight: key recommendations. Available at: http://www.nhlbi.nih.gov/health/public/heart/obesity/lose_wt/recommen.htm. Accessed July 22, 2005.
31. American Heart Association. Be smart for your heart. Available at: http://www.americanheart.org/presenter.jhtml?identifier=502. Accessed July 22, 2005.
32. Serdula MK, Khan LK, Dietz WH. Weight loss counseling revisited. JAMA. 2003;289:1747–1750.
33. Samaha FF, Iqbal N, Seshadri P, et al. A low-carbohydrate as compared with a low-fat diet in severe obesity. N Engl J Med. 2003;348:2074–2081.
34. Foster GD, Wyatt HR, Hill JO, et al. A randomized trial of a low carbohydrate diet for obesity. N Engl J Med. 2003;348:2082–2090.
35. Nilsson PM, Nilsson JA, Hedblad B, Berglund G, Lingarde F. The enigma of increased non-cancer mortality after weight loss in healthy men who are overweight or obese. J Intern Med. 2002;252: 70–78.[CrossRef][ISI][Medline]
36. Wedick NM, Barett-Connor E, Knoke JD, Wingard DL. The relationship between weight loss and all-cause mortality in older men and women with and without diabetes mellitus: the Rancho Bernardo Study. J Am Geriatr Soc. 2002;50:1810–1815.[CrossRef][ISI][Medline]
37. Sorenson TIA. Weight loss causes increased mortality: pros. Obes Rev. 2003;4:3–7.[CrossRef][Medline]
38. Gregg EW, Gerzoff RB, Thompson TJ, Williamson DF. Intentional weight loss and death in overweight and obese US adults 35 years of age and older. Ann Intern Med. 2003;138:383–390.
39. Stampfer MJ, Hu FB, Manson JE, Rimm EB, Willett WC. Primary prevention of coronary heart disease in women through diet and lifestyle. N Engl J Med. 2000;343:16–22.
40. Stefanick ML, Mackey S, Sheehan M, Ellsworth N, Haskell WL, Wood PD. Effects of diet and exercise in men and postmenopausal women with low levels of HDL cholesterol and high levels of LDL cholesterol. N Engl J Med. 1998;339:12–20.
This article has been cited by other articles:
|
|
 |
|
  B. D. Wagner, G. K. Grunwald, J. S. Rumsfeld, J. O. Hill, P. M. Ho, H. R. Wyatt, and A. L. W. Shroyer Relationship of Body Mass Index With Outcomes After Coronary Artery Bypass Graft Surgery Ann. Thorac. Surg., July 1, 2007; 84(1): 10 - 16. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Feinglass, S. Lin, J. Thompson, J. Sudano, D. Dunlop, J. Song, and D. W. Baker Baseline Health, Socioeconomic Status, and 10-Year Mortality Among Older Middle-Aged Americans: Findings From the Health and Retirement Study, 1992 2002 J. Gerontol. B. Psychol. Sci. Soc. Sci., July 1, 2007; 62(4): S209 - S217. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
