Family Practice Vol. 19, No. 3, 272-277
© Oxford University Press 2002
Long-term mortality assessment using biological measures among elderly people. Ten-year follow-up of 597 healthy elderly subjects in Taiwan
Department of Internal Medicine, College of Medicine,
a Institute of Preventive Medicine, College of Public Health, National Taiwan University, Taipei and
b Community Medicine Research Center and Institute of Public Health, National Yang-Ming University, Taipei, Taiwan.
Dr Tony Hsiu-Hsi Chen, Institute of Preventive Medicine, College of Public Health, National Taiwan University, Taipei, Taiwan.
Wu T-H, Lee T-K, Yen M-F, Tung T-H and Chen TH-H. Long-term mortality assessment using biological measures among elderly people. Ten-year follow-up of 597 healthy elderly subjects in Taiwan. Family Practice 2002; 19: 272–277.
Received 1 May 2001; Revised 22 October 2001; Accepted 7 January 2002.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionAppendixReferences |
|---|
Background. Identifying biological measures that are predictive of mortality for elderly people aged over 65 years has not been fully elucidated in oriental studies.
Objective. The associations between these biological measures and long-term mortality were therefore investigated, and classifications for risk of death were developed among Taiwanese elderly people
Methods. Data used in this study were derived from a total of 597 apparently healthy subjects aged over 65 years identified from a nationwide survey that was conducted between 1989 and 1991 in Taiwan. Each participant received a physical examination and a wide range of biological measures. These 597 apparently healthy subjects were followed to 31 December 1999 to determine the cause of death. The grouping technique using factor analysis was first used to aggregate similar characteristics of biological measures into reduced components. Risk of death for each subject was classified into four groups: good (A), fair (B), modest (C) and poor (D). Hazard ratios for groups B, C, D against A were calculated.
Results. The overall 10-year survival rate was 72% [95% confidence interval (CI) 68–76%]. The adjusted hazard ratios for all-cause death in high and mid-level categories of haematological components versus the lowest group were 0.51 (95% CI 0.33–0.80) and 0.56 (95% CI 0.37–0.85), respectively. Group D had a 6-fold risk of death as compared with group A (relative risk = 6.34, 95% CI 3.85–10.52). The corresponding figures were 2.48 (95% CI 1.43–4.29) and 1.60 (95% CI 2.88–6.89) for groups B and C, respectively.
Conclusions. The relationships of biological measures to long-term mortality were elucidated. Information on classification for risk of death may be helpful for elderly people to pay attention to their health status after receiving a health check-up.
Keywords. Biological measures, elderly people, long-term mortality.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
As the population ages, there is an even greater demand for elderly people over 65 years of age to take regular health check-ups. Apart from regular physical examinations, biological measures have been collected to aid physicians in making differential diagnoses. Such measures include carbohydrate, protein, electrolytes, serum enzyme activities, lipids, urinary factors and haematological components. Although the usefulness of biological measures in relevant clinical fields has been recognized, their relationships to long-term mortality for apparently healthy elderly people have not yet been fully addressed. In addition, although the effects of certain biological factors were assessed, inconsistent results were reported. Campbell et al.1 and Stevens et al.2 found negative associations between haemoglobin levels and all-causes mortality. However, neither of these findings were found in the van Asperen et al. study.3
There are two other shortcomings in earlier studies. First, previous studies on this subject were based primarily on clinical patients rather than healthy elderly people. The weakness of such a design is that the impact of laboratory abnormalities may be confounded by co-existing morbidity. The second drawback was that each of them merely addressed the relationship of a single biological factor to mortality, but there was a lack of studies that aim to predict mortality for the elderly based upon a variety of comprehensive biological measures accrued from health check-ups.
The aim of this study was therefore to identify key modifiable biological markers that were predictive of long-term mortality for healthy elderly Taiwanese people aged 65 years and over.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
Study design and sample selection
A nationwide study of elderly people aged 65 years or older was conducted between 1989 and 1991 in Taiwan. Based on a stratified random sample design, subjects were sampled from four cities including Taipei, Taichung, Kaohsiung and Hualien, representative of the northern, central, southern and eastern areas of Taiwan, respectively. For each city, all subjects 65 years and older were listed according to the smallest geographical administrative unit in Mandarin Chinese, called a ‘Li’. Two samples were then selected randomly from each ‘Li’ among the four cities. A total of 2600 subjects (1322 males and 1278 females) participated in this study. Each participant underwent a physical examination and a variety of biological measures.
A general physical examination was administered initially by a physician, covering a wide range of principal organs such as skin, eyes, ear, tongue, and so on. Subjects with any suspected disease were referred to a medical centre to receive confirmatory diagnoses. We classified these diseases into seven groups: cardiovascular disease, diabetes mellitus, cerebrovascular disease, respiratory disease, disease of the digestive system, disease of the urinary system and others. Participants classified into one of the disease categories were excluded from further analysis. A total of 597 apparently healthy subjects were left to form the main data set for analysis.
This cohort was followed until the end of December 1999 to ascertain the cause of death. Biological measures collected in this study include:
- Carbohydrate: glucose and fructosamine.
- Proteins: total protein, albumin and globulin.
- Lipids: total cholesterol (CHO), high-density lipoprotein (HDL), low-density lipoprotein (LDL) and triglyceride (TG).
- Non-protein nitrogenous compounds: urea nitrogen (UN), creatinine (CRE) and uric acid (UA).
- Serum enzymes: r-glutamyl transferase (r-GT), lactic dehydrogenase (LDH), glutamic–oxyacetic transaminase (GOT), glutamic–pyvuvic transaminase (GPT) and creatine phosphokinase (CPK).
- Electrolytes [including potassium (K), sodium (Na), magnesium (Mg), calcium (Ca), iron (IRN) and chloride (Cl)].
- Haematology: red blood cells (RBC), haemoglobin (Hb), haematocrit (Ht) and white blood cells (WBC).
- Others: bilirubin.
Statistical method
As biological markers may interact with complexity, the grouping technique using factor analysis was first used to combine similar characteristics of biological measures, RBC and Hb for example, into reduced components. This technique enables one to identify the key modifiable biological markers that were predictive of mortality with parsimoniousness. Table 1
shows the contributory weight (factor coefficients) from each biological measure to the corresponding new component. The factor score for each new component was calculated by multiplying these weights by the corresponding values of biological measures.4,5 Risk scores based on the estimated regression coefficients using a proportional hazard model were derived further to classify individuals into four groups: good, fair, modest and poor.6,7 A detailed procedure for calculating risk scores is shown in the Appendix. The Kaplan–Meier estimate of cumulative survival among the four groups was compared and tested. Hazard ratios for risk of death for the poor, modest and fair groups against the good group were also calculated using a proportional hazard model.
|
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
Reducing similar characteristics of biological measures into new components
Table 1
shows the results of factor coefficients according to biological measures for each new and reduced component. These include haematology, serum enzyme activities, Na + Cl, urinary chemistries, CHO + LDL, protein, bilirubin, carbohydrate, HDL + K, ALP + WBC, and IRN + Mg. The larger the factor coefficients, the more its biological measures contribute to the new component. One could use these factor coefficients to calculate an overall factor score for each new component.
Overall survival and the association between biological measures and mortality
The median follow-up as of 31 December 1999 was 10 years. Six major causes of death were identified, including cardiovascular disease (CVD, ICD: 430–438); diabetes mellitus (DM, ICD: 250); cerebrovascular disease (CV, ICD: 410–414); respiratory disease (CHE, ICD: 490–496); disease of the digestive system (GI, ICD: 530–537); cancer (ICD: 140–239) and others. A total of 169 deaths were identified among 597 subjects, which include 22 deaths from CV, 75 from CVD, eight from DM, 15 from DCHE, 10 from DGI, 34 from cancer and five from other causes. The overall 5- and 10-year survival rates were 0.91 [95% confidence interval (CI) 89–93%] and 0.72 (95% CI 68–76%). Table 2 shows adjusted hazard ratios for the association between all-cause death and the tertile distribution of new components after controlling for age, sex, body mass index (BMI) and blood pressure. The highest category of haematological components can decrease the risk of death by half as compared with the lowest group after adjustment for age, sex, BMI and blood pressure. This relationship reaches statistical significance. Inverse relationships were also found for electrolytes, CHO + LDL, bilirubin and proteins. Positive associations were noted for serum enzymes activities, urinary chemistries, carbohydrate and ALP + WBC. However, only the haematological component, HDL + K, the highest tertile distribution of protein and ALP + WBC show statistical significance, and no significant associations have been found for other components.
|
The classification of risk of death
According to the quartile distribution of risk score (shown in the Appendix), 156 subjects were classified into the good group (A); 147 in the fair group (B); 147 in the modest group (C); and 147 in the poor group (D). Figure 1
shows significant differences of cumulative survival amongst the four groups, with 10-year survival rates as follows: group A 87.3% (95% CI 81.7–92.8%); group B 80.9% (95% CI 74.2–87.6%); group C 73.3% (95% CI 66.0–80.7%); and group D 46.8% (79 deaths, 95% CI 38.2–55.4%). The differences of cumulative survival reach statistical significance (log-rank chi-square (3) = 92.90, P < 0.0001).
|
Table 3
shows hazard ratios of the fair group, modest group and poor group versus the good group. Group D had a 6-fold risk of death compared with group A (HR = 6.34, 95% CI 3.85–10.52). The corresponding figures were 2.48 (95% CI 1.43–4.29) and 1.60 (95% CI 2.88–6.89) for groups B and C, respectively.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
Clinical implications
The association between biological measures and long-term mortality among subjects 65 years and over was performed on the basis of 597 healthy subjects identified from a community-based survey. The significant finding in our study was that the haematological component could reduce the risk for either all-cause death or specific-cause death by half. Our results were in agreement with the findings of Campbell1 and Stevens,2 which reported the negative association between levels of haemoglobin and mortality due to all causes and cancer. It may be postulated that the mechanism accounting for such a negative association between levels of haematological components and mortality was that high haemoglobin levels may lead to an increase in total iron-binding capacity, and, in turn, reduce the body's iron store. Such dynamics were postulated likewise by Stevens et al.8 as a risk factor determining mortality and incidence of cancer.9 In addition to cancer, it was also stipulated by Sullivan10,11 that higher incidences of CVD among men compared with women were attributed to higher levels of iron stores in men. However, the veracity of such a postulate still remains unresolved due to inconsistent results produced by other epidemiological studies on this subject. For instance, no association between body iron stores and cancer was found among Finnish men.12 Three additional studies found a positive relationship between body stored iron and CVD,13–15 whereas two others found an inverse association.12–16 No association was found in the van Asperen et al. study.3 Besides the relationship between mortality and all causes, the negative association between haematological components and CVD was also found in this study (data not shown). Regrettably, we cannot assess the association between haematological components and mortality from cancer due to a sparse number of cases. The real cause accounting for the negative association between haematological components and mortality due to all causes or cause-specific elements such as CVD should be corroborated in future studies.
Other significant findings were the highest tertile of protein, HDL + K and ALP + WBC. Surprisingly, HDL and K were aggregated as the same component and demonstrated as a protective factor for long-term mortality. This may be attributed to the fact that both were reported to be significantly associated with the risk of heart disease. This postulate was corroborated by the finding that a negative relationship between HDL + K and mortality from CVD was also found with estimates of the hazard ratios, 0.67 (95% CI 0.22–2.05) and 0.17 (95% CI 0.04–0.66), for the middle and highest group against the lowest group. It is very interesting to see that the highest tertile of protein may significantly reduce the risk of death. As the highest tertile of protein can significantly reduce the risk of deaths from CVD (0.20; 95% CI 0.05–0.87), CVD (0.23; 95% CI 0.07–0.77) and other diseases, this suggests that protein component may play an important role in overall health status for elderly people. However, the exact biological mechanism should be explored in the future. The detrimental result of highest tertile of ALP + WBC was consistent with previous findings that a higher WBC has been demonstrated to be a risk factor for myocardial infarction, ischaemic heart disease and CVD.17–19
Risk classification
Risk of all-cause death can be classified on the basis of the above associations between the quartile distribution of each factor score and long-term mortality. The proposed method can be easily accommodated to other data (the program is available from Tony H-H Chen on request). The risk scores derived from these relationships were used to classify subjects into four groups, ranging from a low-risk group (group A) to one of high risk (group D). Group D was found to be six times more likely to die as compared with group A. The results of these findings not only aid physicians in the evaluation of the subject's prognosis, but also prompt elderly people to pay heed to their own health status after regular health check-ups.
Methodological consideration
In order to assess whether the associations between biological measures and mortality were valid using our proposed model, split-sample validation techniques were performed to divide the 597 subjects into two equal samples: the trained sample and the validation sample. Trained samples were used to obtain regression coefficients in the Cox regression model, as exemplified above. These parameters were then applied to the validation sample to calculate the predicted survival. As the predicted survival curve was close to the observed (Figure 2) (r2 = 0.97), this suggests that our proposed model of mortality assessment through the use of biological measures is valid.
|
In spite of this, there is one limitation in this study. Although our proposed model can be applied similarly to classify the risk(s) of death from specific causes, the number of deaths for each specific cause of death was low, and thus precluded us from elucidating the associations between biological measures and specific-cause mortality. A long-term follow-up or other large studies may be required to provide exhaustive mortality assessment for each specific cause.
In conclusion, the present study used a community-based cohort to identify the key modifiable biological markers that were predictive of long-term mortality among healthy elderly Taiwanese people aged 65 years and over. These relationships were used further to categorize individuals into different groups according to risks of death. Information provided in this way may be useful for elderly people to pay attention to their own health status.
|
Appendix |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
According to the Cox regression model, the equation for hazard rates of death associated with the tertile distribution of factor scores was expressed as:
|
|
The estimated coefficients from the expression (A-1) yield the following risk score:
|
|
Based on the quartile distribution of S, four groups were derived as follows:
- Good S<A (529.07)
- Fair A
S<B (592.64)
- Modest BS<C (650.43)
- Poor S
C
|
Acknowledgments |
|---|
This study was supported by a grant (DOH80-25) from the Department of Health, Executive Yuan, Taiwan.
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendixReferences |
|---|
1 Campbell MJ, Elwood PC, Mackean J, Waters WE. Mortality, hemoglobin level and haematocrit in women. J Chron Dis 1985; 38: 881–889.[Web of Science][Medline]
2
Stevens RG, Kuvibidila S, Kapps M, Friedlaender J, Blumberg BS. Iron-binding proteins, hepatitis B virus, and mortality in the Solomon Islands. Am J Epidemiol 1983; 118: 550–561.
3
van Asperen IA, Feskens EJM, Bowles CH, Kromhout D. Body iron stores and mortality due to cancer and ischaemic heart disease: a 17-year follow-up study of elderly men and women. Int J Epidemiol 1995; 24: 665–670.
4 Kleinbaum DG, Kupper LL, Muller KE. Applied Regression Analysis and Other Multivariable Methods, 2nd edn. PWS-KENT Publishing Company, 1988.
5 Sharma S. Applied Multivariate Techniques. John Wiley & Sons, Inc., 1996.
6 SAS Institute, Inc. SAS/STAT User's Guide, Version 6. 4th edn. Vol. 1–2. Cary (NC): SAS Institute Inc., 1990.
7 SAS Institute, Inc. SAS/STAT Software: Changes and Enhancements, through Release 6.11 (55356). Cary (NC): SAS Institute Inc., 1990.
8 Stevens RG, Beasley RP, Blumberg BS. Iron-binding proteins and risk of cancer in Taiwan. J Natl Cancer Inst 1986; 76: 605–610.
9 Stevens RG, Jones DY, Micozzi MS, Taylor PR. Body iron stores and the risk of cancer. N Engl J Med 1988; 319: 1047–1052.[Abstract]
10 Sullivan JL. Iron and the sex difference in heart disease. Lancet 1981; I: 1293–1294.
11 Sullivan JL. The iron paradigm of ischaemic heart disease. Am Heart J 1989; 117: 1177–1188.[Web of Science][Medline]
12 Takkunen H, Reunanen A, Knekt P, Aromaa A. Body iron stores and the risk of cancer [letter]. N Engl J Med 1989; 320: 1014.[Medline]
13
Salonen JT, Nyyssonen K, Korpela H, Tuomilehto J, Seppanen R, Salonen R. High stored iron levels are associated with excess risk of myocardial infarction in Eastern Finnish men. Circulation 1992; 86: 803–811.
14 Morrison HI, Semenciw RM, Mao Y, Wigle DT. Serum iron and risk of fatal acute myocardial infarction. Epidemiology 1994; 5: 243–246.[Web of Science][Medline]
15
Magnusson MK, Sigfusson N, Sigvaldason H, Gudmundur MJ, Magnusson S, Thorgeirsson G. Low iron-binding capacity as a risk factor for myocardial infarction. Circulation 1994; 89: 102–108.
16
Sempos CT, Looker AC, Gillum RF, Makuc DM. Body iron stores and the risk of coronary heart disease. N Engl J Med 1994; 330: 1119–1124.
17
Grimm RH Jr, Neaton PD, Ludwig W. Prognostic importance of the white blood cell for coronary, cancer, and all cause mortality. J Am Med Assoc 1985; 254: 1932–1937.
18
Yarnell JW, Baker IA, Sweetnam PM. Fibrinogen, viscosity, and white blood cell count are major risk factors for ischemic heart disease: the Caerphilly and Speedwell collaborative heart disease studies. Circulation 1991; 83: 836–844.
19 Friedman GD, Klatsky AL, Siegelaub AB. The leukocyte count as a predictor of myocardial infarction. N Engl J Med 1974; 290: 1275–1278.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  H.-K. Chen, C.-D. Tseng, S.-C. Wu, T.-K. Lee, and T. H.-H. Chen A prospective cohort study on the effect of sexual activity, libido and widowhood on mortality among the elderly people: 14-year follow-up of 2453 elderly Taiwanese Int. J. Epidemiol., October 1, 2007; 36(5): 1136 - 1142. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||