Albumin-Corrected Fructosamine Predicts All-Cause and Non-CVD Mortality Among the Very Elderly Aged 80 Years or Older Without Diabetes

Abstract

Background

Several guidelines have suggested alternative glycemic markers for hemoglobin A1c among older adults with limited life expectancy or multiple coexisting chronic illnesses. We evaluated associations between fructosamine, albumin-corrected fructosamine (AlbF), fasting plasma glucose (FPG), and mortality in the diabetic and nondiabetic subpopulations, and compared which marker better predicts mortality among participants aged 80 and older.

Methods

Included were 2 238 subjects from the Healthy Ageing and Biomarkers Cohort Study (2012–2018) and 207 participants had diabetes at baseline. Multivariable Cox proportional hazards regression models investigated the associations of fructosamine, AlbF, FPG, and all-cause, cardiovascular disease (CVD), and non-CVD mortality in the diabetic and nondiabetic subpopulations. Restricted cubic splines explored potential nonlinear relations. C-statistic, integrated discrimination improvement (IDI), and net reclassification improvement (NRI) evaluated the additive value of different glycemic markers to predict mortality.

Results

Overall, 1 191 deaths were documented during 6 793 person-years of follow-up. In the linear model, per unit increases of fructosamine, AlbF, and FPG were associated with a higher risk of mortality in nondiabetic participants, with hazard ratios of 1.02 (1.00, 1.05), 1.27 (1.14, 1.42), and 1.04 (0.98, 1.11) for all-cause mortality, and 1.04 (1.00, 1.07), 1.38 (1.19, 1.59), and 1.10 (1.01, 1.19) for non-CVD mortality, respectively. Comparisons indicated that AlbF better predicts all-cause and non-CVD mortality in nondiabetic participants with significant improvement in IDI and NRI.

Conclusions

Higher concentrations of fructosamine, AlbF, and FPG were associated with a higher risk of all-cause or non-CVD mortality among the very elderly where AlbF may constitute an alternative prospective glycemic predictor of mortality.

Hemoglobin A1c (HbA1c) levels have been widely applied to monitor long-term glycemic status. However, HbA1c monitoring is unstable for people with any condition affecting erythrocyte life span, such as chronic kidney disease (CKD), pregnancy, or certain cancers (1). The findings of several studies along with guideline recommendations revealed that other glycemic markers may be essential to predict short-term risks (eg, mortality) among diabetic patients with limited life expectancy (2–4). The American Diabetes Association also recommends the use of plasma blood glucose or fingerstick readings for glycemic level assessment among older adults with coexisting illness and compromised functional status (1). Considering the potential limitations of HbA1c, fructosamine (also known as glycated serum protein) has gradually been considered as a glycemic marker for the diagnosis of diabetes or a predictor of microvascular complications. Fructosamine is a result of glycation reactions between glucose and serum proteins, most of which involve albumin (typically 90%) in the blood circulation (5–8). The reaction rate and the percentage of glycated protein are dependent on both blood glucose and protein concentrations, because minor fluctuations in albumin can affect fructosamine levels. Hence, the use of protein-corrected serum fructosamine levels (usually measured as albumin-corrected fructosamine [AlbF]) rather than uncorrected data has been proposed as a reliable marker of glycemic control (8–12). Unlike HbA1c, whose half-life is 8–12 weeks, fructosamine reflects the glycemic status of a person over 1–3 weeks influenced by the half-life time of serum proteins. Although fructosamine and AlbF might be influenced by renal or liver disease that changes the rates of production or disappearance (ie, the half-life time) of albumin and other serum proteins, they are not affected by erythrocyte life span, blood loss or transfusion, hemoglobin variants, anemia, or iron deficiency (1,8,10,13,14).

Our previous studies revealed that coexisting illnesses like anemia and CKD are common among the very elderly (≥80 years) (15,16). Hence, HbA1c may not be an effective and stable glycemic marker in this population. Research on fructosamine revealed comparable performance to HbA1c in predicting diabetes, CKD, and retinopathy in the community (13,17), yet the performance of fructosamine in predicting mortality in diabetic or nondiabetic subpopulations remains unclear. Limited evidence indicated that lower AlbF was related to higher all-cause and non-cardiovascular disease (CVD) mortality in patients with diabetic peritoneal dialysis (18), yet the consistency of results needs to be confirmed in other studies. U-shaped associations between HbA1c and mortality may exist among the very elderly (9,18), but whether an association occurs between AlbF and mortality among this population is still unknown. Moreover, the performance of fructosamine or AlbF in comparison to standard glycemic markers as indicators of mortality among the very elderly needs to be further explored.

In this study, we recruited 2 238 very elderly participants from the Healthy Ageing and Biomarkers Cohort Study (HABCS) to investigate the associations of fructosamine and AlbF with mortality in conditions where HbA1c may not be appropriate. Fasting plasma glucose (FPG) was also measured for comparison. Restricted cubic splines were applied to explore the dose–response associations of AlbF, fructosamine, and FPG with all-cause, CVD, and non-CVD mortality. Additional values of different glycemic markers in predicting mortality were also compared.

Method

Study Population

The study used data from the HABCS (19), a subcohort of the Chinese Longitudinal Healthy Longevity Survey (20,21). HABCS was conducted in 9 longevity locations to explore determinants of longevity. The detailed study design has been previously described (19,20). Overall, we identified 3 568 older adults investigated in the 2012 and 2014 baselines who were potential candidates for analysis. Individuals aged younger than 80 years (n = 1 106), with missing values for fructosamine or albumin (n = 141), with albumin values less than 30 g/L (n = 70), and with uncertain death time (n = 13) were excluded, leaving 2 238 very elderly (mean age: 92.7 ± 7.7 years) in the final cohort who were successfully followed up to 2018 (Supplementary Figure 1). The study was approved by the ethics committee of the National Institute of Environmental Health, Chinese Center for Disease Control and Prevention (No. 201922), and all participants or their proxy respondents provided informed consent.

Measurement of Exposures

For each participant, 7 mL of venous blood was collected for biochemical determination. Fructosamine and FPG were analyzed by an Automatic Biochemistry Analyzer (Hitachi 7180, Japan) using commercial diagnostic kits (Roche Diagnostic, Mannheim, Germany). For fructosamine, the measured range was 10–1 000 μmol/L, with an interassay coefficient of variation of 2.9% at 296 μmol/L. FPG was reported as mmol/L (1 mmol/L = 18 mg/dL). Plasma albumin was determined using the bromocresol green method. AlbF was calculated by fructosamine (μmol/L)/plasma albumin (g/L) (9,18). Participants with plasma albumin less than 30 g/L were excluded based on the work of Welsh et al. (8).

Ascertainment of Deaths

For each deceased participant, the date of death was collected from the participant’s family members and verified by qualified doctors. Follow-up time was determined from the baseline interview until either death, loss to follow-up, or the end of follow-up (September 1, 2018), whichever came first. Lost to follow-up was defined by a participant not responding to the follow-up survey (19,22). Cause-specific mortality (cardiovascular or noncardiovascular) was collected by a supplementary phone interview in the 2014 survey and assessed by qualified doctors in the 2017 survey. Detailed causes of death for 872 of 1 191 (73.2%) participants were successfully obtained. The 10th version of the International Classification of Disease (ICD-10) was applied to determine the causes of death. ICD-10 codes I00–I99 were assigned to cardiovascular death (23), other causes of death were grouped as noncardiovascular death.

Covariates

Information on individual characteristics, lifestyles, self-reported diseases, and functional status was acquired using questionnaires through household surveys. These variables include age, sex, race, residence, living arrangement, education level, marital status, income levels, insurance, smoking, alcohol drinking, exercise, self-reported cerebrovascular disease, heart disease, respiratory diseases, and cancer; dietary diversity, cognitive impairment, activities of daily living (ADL) disability, visual function, psychological resources, and pain were also evaluated based on questionnaires (24–28). Physical examination was completed by experienced physicians to acquire systolic blood pressure, diastolic blood pressure, height, and weight to ascertain hypertension and body mass index (BMI) (29,30). Laboratory tests were performed to determine serum creatinine, superoxide dismutase (SOD), malondialdehyde (MDA), plasma 25(OH)D, vitamin B12, hemoglobin (Hb), high-sensitivity C-reactive protein (hs-CRP), total cholesterol (TC), triglyceride (TG), high-density lipoprotein (HDL), uric acid (UA), and blood urea nitrogen (BUN) as described before (19,22,31). Diabetes was diagnosed using FPG, self-reported prior diagnosis, or related medication use. Anemia and CKD were diagnosed using Hb and estimate glomerular filtration rate measurements, respectively (32–34). For detailed definitions, see Supplementary Methods.

Statistical Analysis

In total, missing data included 8.1% (182/2 238) of BMI readings but less than 4.5% in any other covariate. A multiple imputation method based on the random forest method was implemented to generate 5 data sets to deal with the potential impacts of missing data on model performance (35,36).

We performed Cox proportional hazards models to explore the relationship between fructosamine, AlbF, FPG, and all-cause, CVD, and non-CVD mortality among the diabetic and nondiabetic subpopulations, respectively. Fructosamine, AlbF, and FPG were considered as continuous variables or quintiles in analyses. All models met the proportional hazards assumption (all p values >.20). To explore the potential nonlinear associations, we replaced missing values with the mean and mode of imputed variables. Hazard ratio (HR) and 95% confidence interval (CI) were estimated in 6 models adjusted for different covariates, in which the crude model was not adjusted for any other variable; Model 1 was further adjusted for age, sex, race, and residence; Model 2 was further adjusted for living arrangement, education level, marital status, income level, and insurance; Model 3 was further adjusted for the current status of smoking, alcohol drinking, exercise, and dietary diversity; Model 4 was further adjusted for BMI, psychological resources, hypertension, heart disease, cerebrovascular disease, anemia, respiratory diseases, CKD, ADL disability, cognitive impairment, visual function, cancer, and pain; Model 5 was further adjusted for TC, TG, HDL, MDA, SOD, plasma 25(OH)D, vitamin B12, BUN, hs-CRP, and UA. We also estimated cumulative mortality via Kaplan–Meier curves and tested the statistical significance of difference in different quintiles of glycemic markers using the log-rank test (37). Effect modification was explored by sex and race. We also conducted a complete-case analysis to assess the influence of the imputation process.

To assess the added values of different glycemic markers on mortality, we compared C-statistic, integrated discrimination improvement (IDI), and net reclassification improvement (NRI) in fully adjusted models with alternative glycemic markers (38,39). The C-statistic measured the consistency between model estimates and observed events. NRI and IDI were calculated between models with AlbF to measure the added predictive values to models without AlbF (“Main model” in result tables), with fructosamine, or with FPG and albumin.

Data were presented as mean ± standard deviation (SD) for continuous variables with normal distribution and compared via one-way analysis of variance, or median (interquartile range [IQR]) for continuous variables with skewed distribution and compared by Kruskal–Wallis tests, and number of participants (percentage) for categorical variables and compared using χ ² tests. All statistical analyses were conducted in R 3.6.2 for Windows (R Foundation for Statistical Computing, Vienna, Austria) with the “rms,” “survival,” “survminer,” “survIDINRI,” “randomForest,” “mice,” and “ggplot2” packages.

Results

Baseline Characteristics

Over 6 793 person-years of follow-up (median follow-up time 3.0 years; IQR: 1.8–4.3), 1 191 deaths were documented, with cause-specific deaths for 872 (73.2%) participants, including 236 for CVD and 636 for non-CVD. Table 1 and Supplementary Table 1 present baseline characteristics by status of diabetes. Participants without diabetes were likely to be older, Han race, underweight, unmarried, divorced or widowed; likely to have anemia; with higher values of HDL and SOD; with lower values of fructosamine, AlbF, FPG, TG, and hs-CRP. The Pearson correlation coefficients between AlbF and fructosamine or FPG were higher in diabetic participants (0.94 and 0.75, respectively) compared to participants without diabetes (0.53 and 0.17, respectively; Supplementary Figure 2).

Table 1.

Main Baseline Characteristics of 2 238 Very Elderly by Diabetes

Characteristics	Diabetes			p
	Yes (n = 207)	No (n = 2 031)	Overall (n = 2 238)
Age (years)	91.5 (7.7)	92.9 (7.7)	92.7 (7.7)	.015
Sex
Male	80 (38.6)	712 (35.1)	792 (35.4)	.341
Female	127 (61.4)	1 319 (64.9)	1 446 (64.6)
Smoking
No	176 (85.0)	1 790 (88.1)	1 966 (87.8)	.184
Yes	27 (13.0)	200 (9.8)	227 (10.1)
Missing	4 (1.9)	41 (2.0)	45 (2.0)
Alcohol drinking
No	180 (87.0)	1 757 (86.5)	1 937 (86.6)	.876
Yes	22 (10.6)	229 (11.3)	251 (11.2)
Missing	5 (2.4)	45 (2.2)	50 (2.2)
BMI
Underweight	46 (22.2)	586 (28.9)	632 (28.2)	.018
Normal	102 (49.3)	978 (48.2)	1 080 (48.3)
Overweight or obese	44 (21.3)	300 (14.8)	344 (15.4)
Missing	15 (7.2)	167 (8.2)	182 (8.1)
Hypertension
No	77 (37.2)	779 (38.4)	856 (38.2)	.802
Yes	130 (62.8)	1 252 (61.6)	1 382 (61.8)
Heart disease
No	183 (88.4)	1 819 (89.6)	2 002 (89.5)	.291
Yes	19 (9.2)	139 (6.8)	158 (7.1)
Missing	5 (2.4)	73 (3.6)	78 (3.5)
Cerebrovascular disease
No	180 (87.0)	1 799 (88.6)	1 979 (88.4)	.258
Yes	21 (10.1)	154 (7.6)	175 (7.8)
Missing	6 (2.9)	78 (3.8)	84 (3.8)
Anemia
No	106 (51.2)	830 (40.9)	936 (41.8)	.008
Yes	99 (47.8)	1 153 (56.8)	1 252 (55.9)
Missing	2 (1.0)	48 (2.4)	50 (2.2)
Respiratory diseases
No	184 (88.9)	1 788 (88.0)	1 972 (88.1)	.794
Yes	17 (8.2)	165 (8.1)	182 (8.1)
Missing	6 (2.9)	78 (3.8)	84 (3.8)
CKD
No	84 (40.6)	919 (45.2)	1 003 (44.8)	.225
Yes	123 (59.4)	1 112 (54.8)	1 235 (55.2)
Cancer
No	201 (97.1)	2 008 (98.9)	2 209 (98.7)	.069
Yes	6 (2.9)	23 (1.1)	29 (1.3)
Use of diabetes medications
No	194 (93.7%)	2 031 (100%)	2 225 (99.4%)	<.001
Yes	13 (6.3%)	0 (0%)	13 (0.6%)
Psychological resources (points)	24.0 (4.50)	24.1 (4.92)	24.1 (4.88)	.612
Fructosamine (10 μmol/L)	28.1 (8.94)	23.9 (2.82)	24.3 (4.01)	<.001
AlbF (μmol/g)	6.97 (2.18)	5.98 (0.60)	6.07 (0.92)	<.001
FPG (mmol/L)	8.93 (3.54)	4.52 (1.08)	4.92 (1.96)	<.001
Plasma albumin (g/L)	40.4 (4.47)	40.2 (4.24)	40.2 (4.27)	.387

Notes: AlbF = albumin-corrected fructosamine; BMI = body mass index; CKD = chronic kidney disease; FPG = fasting plasma glucose. Data are presented as mean (SD), median (IQR), or number of participants (percentage).

All-Cause Mortality

In the fully adjusted linear model, per unit increases of fructosamine, AlbF, and FPG corresponded to 5%, 24%, and 7% higher risk of all-cause mortality in diabetic patients, respectively (Table 2). While in nondiabetic participants, only AlbF was found to be associated with a higher risk of all-cause mortality, with an HR of 1.27 (1.14, 1.42; Table 3). When compared to the lowest quintile of glycemic markers in diabetic or nondiabetic participants, statistically significant differences of all-cause mortality were not observed in other quintiles of fructosamine or FPG in diabetic patients, while nondiabetic participants in the fifth quintile of AlbF had a 35% higher risk of all-cause mortality (Tables 2 and 3). These results remained stable when adjusted for different variables in Models 1–5 (Supplementary Tables 2 and 3), when the complete data set was analyzed (Supplementary Tables 4–7), and when further adjusted for FPG in analyses of fructosamine or AlbF (Supplementary Tables 8 and 9). Sex and race did not modify the effects.

Table 2.

Hazard Ratio (95% CI) for All-Cause Mortality, CVD Mortality, and Non-CVD Mortality According to Fructosamine, AlbF, and FPG in the Fully Adjusted Model (Model 5^†, n = 207, with diabetes)

Quintiles	All-Cause Mortality	CVD Mortality	Non-CVD Mortality
Fructosamine (10 μmol/L)
<22.60	Reference	Reference	Reference
22.60–24.66	1.13 (0.59–2.18)	0.01 (0.00–0.56)*	0.90 (0.35–2.30)
24.67–26.97	0.95 (0.48–1.91)	0.38 (0.01–9.49)	0.86 (0.34–2.13)
26.98–30.11	1.19 (0.61–2.35)	0.66 (0.06–7.57)	0.97 (0.37–2.50)
≥30.12	1.97 (0.96–4.03)	0.70 (0.04–12.37)	1.61 (0.58–4.41)
Linear model	1.05 (1.02–1.08)**	1.02 (0.93–1.12)	1.04 (1.00–1.08)
AlbF (μmol/g)
<5.80	Reference	Reference	Reference
5.80–6.10	1.25 (0.62–2.51)	2.20 (0.20–24.19)	0.97 (0.36–2.61)
6.11–6.62	1.01 (0.49–2.11)	2.24 (0.11–44.75)	0.84 (0.30–2.37)
6.63–7.23	1.56 (0.74–3.27)	1.33 (0.05–35.86)	1.32 (0.47–3.70)
≥7.24	2.14 (0.96–4.79)	0.91 (0.06–14.27)	1.97 (0.61–6.35)
Linear model	1.24 (1.11–1.37)**	1.22 (0.85–1.74)	1.17 (1.00–1.36)
FPG (mmol/L)
<7.26	Reference	Reference	Reference
7.26–7.82	0.55 (0.29–1.07)	19.67 (0.40–975.40)	0.45 (0.18–1.13)
7.83–8.53	0.84 (0.43–1.67)	0.57 (0.01–40.82)	0.55 (0.22–1.39)
8.54–9.83	0.51 (0.26–1.03)	0.08 (0.00–1.65)	0.23 (0.08–0.71)*
≥9.84	0.96 (0.48–1.90)	6.23 (0.27–145.02)	0.94 (0.37–2.41)
Linear model	1.07 (1.01–1.15)*	0.87 (0.65–1.15)	1.04 (0.94–1.15)

Notes: AlbF = albumin-corrected fructosamine; FPG = fasting plasma glucose; BMI = body mass index; CKD = chronic kidney disease; ADL = activities of daily living; TC = total cholesterol; HDL = high-density lipoprotein; TG = triglycerides; MDA = malondialdehyde; SOD = superoxide dismutase; BUN = blood urea nitrogen; hs-CRP = high-sensitivity C-reactive protein; UA = blood uric acid; CVD = cardiovascular disease; CI = confidence interval.

†Adjusted for age, sex, race, residence, living arrangement, education level, marital status, income level, insurance, smoking, alcohol drinking, exercise, dietary diversity, BMI, hypertension, HD, cerebrovascular disease, anemia, respiratory diseases, CKD, cancer, psychological resources, ADL disability, cognitive impairment, visual function, pain, TC, TG, HDL, MDA, SOD, plasma 25(OH)D, vitamin B12, BUN, hs-CRP, and UA. Albumin was further adjusted if the exposure is FPG.

*p < .05, **p < .001.

Table 3.

Hazard Ratio (95% CI) for All-Cause Mortality, CVD Mortality, and Non-CVD Mortality According to Fructosamine, AlbF, and FPG in the Fully Adjusted Model (Model 5^†, n = 2 031, without diabetes)

Quintiles	All-Cause Mortality	CVD Mortality	Non-CVD Mortality
Fructosamine (10 μmol/L)
<21.60	Reference	Reference	Reference
21.60–23.03	0.93 (0.77–1.12)	0.79 (0.51–1.22)	1.06 (0.81–1.38)
23.04–24.50	0.97 (0.79–1.17)	0.82 (0.52–1.30)	1.19 (0.91–1.56)
24.51–26.15	1.07 (0.87–1.30)	1.13 (0.72–1.75)	1.21 (0.91–1.60)
≥26.16	1.14 (0.93–1.40)	1.10 (0.69–1.73)	1.24 (0.93–1.66)
Linear model	1.02 (1.00–1.05)	1.01 (0.95–1.07)	1.04 (1.00–1.07)*
AlbF (μmol/g)
<5.51	Reference	Reference	Reference
5.51–5.79	0.92 (0.74–1.13)	0.90 (0.58–1.41)	0.92 (0.69–1.22)
5.80–6.05	1.17 (0.96–1.44)	1.34 (0.87–2.06)	1.10 (0.83–1.45)
6.06–6.37	1.14 (0.92–1.40)	0.92 (0.58–1.47)	1.17 (0.89–1.56)
≥6.38	1.35 (1.09–1.66)**	1.24 (0.78–1.97)	1.47 (1.10–1.96)**
Linear model	1.27 (1.14–1.42)***	0.99 (0.76–1.29)	1.38 (1.19–1.59)***
FPG (mmol/L)
<3.72	Reference	Reference	Reference
3.72–4.32	1.10 (0.91–1.34)	1.20 (0.79–1.84)	1.09 (0.83–1.43)
4.33–4.77	1.01 (0.83–1.24)	0.80 (0.50–1.28)	1.06 (0.80–1.41)
4.78–5.35	1.18 (0.97–1.45)	1.27 (0.82–1.98)	1.36 (1.03–1.79)*
≥5.36	1.20 (0.98–1.48)	1.21 (0.77–1.91)	1.38 (1.05–1.83)*
Linear model	1.04 (0.98–1.11)	1.00 (0.88–1.15)	1.10 (1.01–1.19)*

Notes: AlbF = albumin-corrected fructosamine; FPG = fasting plasma glucose; BMI = body mass index; CKD = chronic kidney disease; ADL = activities of daily living; TC = total cholesterol; HDL = high-density lipoprotein; TG = triglycerides; MDA = malondialdehyde; SOD = superoxide dismutase; BUN = blood urea nitrogen; hs-CRP = high-sensitivity C-reactive protein; UA = blood uric acid; CVD = cardiovascular disease; HR = hazard ratio; CI = confidence interval.

†Adjusted for age, sex, race, residence, living arrangement, education level, marital status, income level, insurance, smoking, alcohol drinking, exercise, dietary diversity, BMI, hypertension, HDL, cerebrovascular disease, anemia, respiratory diseases, CKD, cancer, psychological resources, ADL disability, cognitive impairment, visual function, pain, TC, TG, HDL, MDA, SOD, plasma 25(OH)D, vitamin B12, BUN, hs-CRP, and UA. Albumin was further adjusted if the exposure is FPG.

*p < .05, **p < .01, ***p < .001.

In diabetic patients, a U-shaped association between FPG with all-cause mortality was observed with the lowest risk at a value of 9.14 mmol/L, whereas nonlinear associations were not observed between fructosamine or AlbF and all-cause mortality (all p values for nonlinearity >.05; Supplementary Figure 3). In nondiabetic participants, a J-shaped association between fructosamine and all-cause mortality was observed with the lowest risk at a value of 232.6 μmol/L, while nonlinear associations were not observed between AlbF or FPG and all-cause mortality (all p values for nonlinearity >.05; Supplementary Figure 4). Moreover, while significant differences in all-cause mortality were not observed by quintiles of glycemic markers among diabetic patients (Supplementary Table 10 and Supplementary Figure 5), higher AlbF quintiles were associated with higher all-cause mortality in nondiabetic participants (Figure 1), with 5-year mortalities of 51.15%, 53.02%, 61.30%, 64.49%, and 71.72% for the first to fifth quintiles of AlbF, respectively (Supplementary Table 11).

Figure 1.

The cumulative incidence of all-cause, CVD, and non-CVD mortality in the very elderly without diabetes. Notes: CVD = cardiovascular disease; AlbF = albumin-corrected fructosamine; FPG = fasting plasma glucose. Quintiles for fructosamine (10 μmol/L) are <21.60, 21.60–23.03, 23.04–24.50, 24.51–26.15, and≥26.16； quintiles for AlbF (μmol/g) are <5.51, 5.51–5.79, 5.80–6.05, 6.06–6.37, and ≥6.38； quintiles for FPG (mmol/L) are <3.72, 3.72–4.32, 4.33–4.77, 4.78–5.35, and ≥5.36.

Compared with fructosamine and FPG, AlbF produced a similar predicted C-statistic of 0.710 (0.702, 0.718) for all-cause mortality but showed significant improvements in IDI and NRI in predicting all-cause mortality in nondiabetic participants (Table 5). Predictive performance was similar in diabetic patients (Table 4).

Table 5.

Prediction Performance on Mortality Based on Main Model With Fructosamine, AlbF, FPG or Not (n = 2 031, without diabetes)

Model	Model Performance*
	C-Statistic (95% CI)	IDI (95% CI)	NRI (95% CI)
All-cause mortality
Main model†	0.707 (0.699, 0.715)	0.005 (0.001, 0.013)	0.067 (0.023, 0.120)
+Fructosamine	0.707 (0.699, 0.716)	0.004 (0.000, 0.011)	0.072(0.000, 0.118)
+AlbF	0.710 (0.702, 0.718)	—	—
+Albumin +FPG	0.709 (0.701, 0.717)	0.003 (0.000, 0.010)	0.004 (0.000, 0.011)
CVD mortality
Main model†	0.730 (0.712, 0.748)	0.000 (−0.001, 0.010)	0.014 (−0.043, 0.105)
+Fructosamine	0.730 (0.712, 0.748)	0.000 (−0.001, 0.006)	0.038 (−0.052, 0.123)
+AlbF	0.730 (0.712, 0.748)	—	—
+Albumin +FPG	0.730 (0.712, 0.748)	0.001 (−0.001, 0.013)	0.107 (−0.130, 0.207)
Non-CVD mortality
Main model†	0.719 (0.708, 0.730)	0.005 (0.000, 0.013)	0.082 (0.011, 0.140)
+Fructosamine	0.719 (0.708, 0.730)	0.004 (0.000, 0.011)	0.093(0.000, 0.150)
+AlbF	0.723 (0.712, 0.733)	—	—
+Albumin +FPG	0.722 (0.711, 0.733)	0.003 (0.000, 0.012)	0.077 (−0.035, 0.133)

Notes: AlbF = albumin-corrected fructosamine; FPG = fasting plasma glucose; BMI = body mass index; CKD = chronic kidney disease; ADL = activities of daily living; TC = total cholesterol; HDL = high-density lipoprotein; TG = triglycerides; MDA = malondialdehyde; SOD = superoxide dismutase; BUN = blood urea nitrogen; hs-CRP = high-sensitivity C-reactive protein; UA = blood uric acid; CI = confidence interval; CVD = cardiovascular disease; IDI = integrated discrimination improvement; NRI = net reclassification improvement.

*C-statistics were calculated in main models and models further adjusted for different glycemic markers, respectively; NRI and IDI were calculated between models with AlbF to measure the added predictive values to models without AlbF, with fructosamine, or with FPG and albumin.

^†Adjusted for age, sex, race, residence, living arrangement, education level, marital status, income level, insurance, smoking, alcohol drinking, exercise, dietary diversity, BMI, hypertension, HD, cerebrovascular disease, anemia, respiratory diseases, CKD, cancer, psychological resources, ADL disability, cognitive impairment, visual function, pain, TC, TG, HDL, MDA, SOD, plasma 25(OH)D, vitamin B12, BUN, hs-CRP, and UA.

Table 4.

Prediction Performance on Mortality Based on Main Model With Fructosamine, AlbF, FPG or Not (n = 207, with diabetes)

Model	Model Performance*
	C-Statistic (95% CI)	IDI (95% CI)	NRI (95% CI)
All-cause mortality
Main model†	0.741 (0.717, 0.766)	0.028(−0.001, 0.057)	0.221 (−0.056, 0.363)
+Fructosamine	0.754 (0.730, 0.778)	0.000 (−0.004, 0.014)	−0.009 (−0.115, 0.162)
+AlbF	0.753 (0.729, 0.778)	—	—
+Albumin +FPG	0.749 (0.724, 0.774)	0.020 (−0.007, 0.045)	0.103 (−0.167, 0.339)
CVD mortality
Main model†	0.942 (0.918, 0.966)	0.000 (−0.001, 0.008)	0.014 (−0.043, 0.105)
+Fructosamine	0.942 (0.917, 0.967)	0.000 (−0.001, 0.006)	0.038 (−0.052, 0.123)
+AlbF	0.945 (0.922, 0.968)	—	—
+Albumin +FPG	0.948 (0.927, 0.968)	0.001 (−0.001, 0.013)	0.107(−0.130, 0.207)
Non-CVD mortality
Main model†	0.785 (0.754, 0.816)	0.008 (−0.007, 0.039)	0.109 (−0.199, 0.375)
+Fructosamine	0.784 (0.752, 0.816)	−0.002 (−0.015, 0.014)	0.028 (−0.174, 0.161)
+AlbF	0.786 (0.753, 0.818)	—	—
+Albumin +FPG	0.785 (0.753, 0.817)	0.008 (−0.026, 0.048)	0.109 (−0.245, 0.359)

Notes: AlbF = albumin-corrected fructosamine; FPG = fasting plasma glucose; BMI = body mass index; CKD = chronic kidney disease; ADL = activities of daily living; TC = total cholesterol; HDL = high-density lipoprotein; TG = triglycerides; MDA = malondialdehyde; SOD = superoxide dismutase; BUN = blood urea nitrogen; hs-CRP = high-sensitivity C-reactive protein; UA = blood uric acid; CI = confidence interval; CVD = cardiovascular disease; IDI = integrated discrimination improvement; NRI = net reclassification improvement.

*C-statistics were calculated in main models and models further adjusted for different glycemic markers, respectively; NRI and IDI were calculated between models with AlbF to measure the added predictive values to models without AlbF, with fructosamine, or with FPG and albumin.

^†Adjusted for age, sex, race, residence, living arrangement, education level, marital status, income level, insurance, smoking, alcohol drinking, exercise, dietary diversity, BMI, hypertension, HD, cerebrovascular disease, anemia, respiratory diseases, CKD, cancer, psychological resources, ADL disability, cognitive impairment, visual function, pain, TC, TG, HDL, MDA, SOD, plasma 25(OH)D, vitamin B12, BUN, hs-CRP, and UA.

CVD Mortality

In the fully adjusted linear model, statistically significant associations were not observed between glycemic markers and CVD mortality (Tables 2 and 3). When compared to the lowest quintile of glycemic markers, statically significant differences involving CVD mortality were only observed in the second quintile of fructosamine in diabetic patients, with an HR of 0.01 (0.00, 0.56; Table 2). The results remained stable when adjusted for different variables in Models 1–5 (Supplementary Tables 2 and 3), when the analysis was conducted on the complete data set (Supplementary Tables 4–7), and when analyses of fructosamine or AlbF were further adjusted for FPG (Supplementary Tables 8 and 9). When AlbF analyses were further adjusted for FPG, per unit increases in AlbF corresponded to a 101% higher risk of CVD mortality in the fully adjusted model among diabetic patients, with an HR of 2.01 (1.11, 3.65; Supplementary Table 8). Sex and race did not modify the effects.

Compared with fructosamine and FPG, AlbF derived a similar C-statistic for CVD mortality and without significant improvement in IDI and NRI in predicting CVD mortality. Predictive performance was similar in diabetic and nondiabetic subpopulations (Tables 4 and 5).

Non-CVD Mortality

In the fully adjusted linear model, per unit increases of fructosamine, AlbF, and FPG corresponded to 4%, 38%, and 10% higher risk of non-CVD mortality in nondiabetic participants, respectively (Table 3). Significant linear associations were not observed in diabetic patients (Table 2). When compared to the lowest quintile of fructosamine, AlbF, or FPG, a significantly lower risk of non-CVD mortalities was observed in the fourth quintile of FPG in diabetic patients with an HR of 0.23 (0.08, 0.71; Table 2) and in the fifth quintile of AlbF in nondiabetic participants with an HR of 1.47 (1.10, 1.96) (Table 3), the fourth and fifth quintiles of FPG in nondiabetic participants with HRs of 1.36 (1.03, 1.79) and 1.38 (1.05, 1.83), respectively (Table 3). The results remained stable when adjusted for different variables in Models 1–5 (Supplementary Tables 2 and 3), when the analysis was conducted using the complete data set (Supplementary Tables 4–7), and when further adjusted for FPG in analyses of fructosamine or AlbF (Supplementary Tables 8 and 9). Sex and race again did not have modification effects.

A U-shaped association between FPG with non-CVD mortality was observed with the lowest risk at a value of 9.57 mmol/L in diabetic patients (Supplementary Figure 3). Nonlinear associations were not observed between fructosamine or AlbF and non-CVD mortality in diabetic patients or nondiabetic participants and between FPG and non-CVD mortality in nondiabetic participants (all p values for nonlinearity >.05; Supplementary Figures 3 and 4). The difference in survival curves was significant but with a lower p value when categorized by AlbF compared to FPG (Figure 1). The 5-year mortalities in nondiabetic participants for first to fifth quintiles of AlbF were 33.84%, 33.73%, 37.69%, 46.68%, and 51.37%, respectively (Supplementary Table 11).

Compared with fructosamine and FPG, AlbF had a similar C-statistic of 0.723 (0.712, 0.733) for non-CVD mortality but showed a significant improvement in IDI and NRI in predicting non-CVD mortality in nondiabetic participants (Table 5). Predictive performance was similar in diabetic patients (Table 4).

Discussion

In this large prospective cohort study of 2 238 very elderly Chinese participants (mean age of 92.7 years), we found that AlbF was significantly associated with higher all-cause mortality in nondiabetic participants; fructosamine, AlbF, and FPG were significantly associated with higher all-cause mortality in diabetic patients and higher non-CVD mortality in nondiabetic participants. Findings between fructosamine, AlbF, and mortality remained robust even after adjustment for FPG. C-statistics were similar when fructosamine, AlbF, or FPG were added to the fully adjusted model. In nondiabetic participants, AlbF showed better performance in predicting all-cause and non-CVD mortality with significant improvements in the relative IDI and NRI. Additional prognostic values of model performance and increased mortality risk of higher AlbF suggested that AlbF may constitute a viable alternative glycemic marker in mortality risk assessment among the very elderly.

The associations of fructosamine, AlbF, and FPG with all-cause, CVD, and non-CVD mortality have rarely been studied, especially among nondiabetic participants aged 80 years and older. In nonlinear analysis, the J-shaped association between fructosamine and all-cause mortality in nondiabetic participants was consistent with a previous study that included 93.1% (10 342/11 104) nondiabetic participants with a mean age of about 57 years, revealing a similar point of lowest risk of mortality at approximately 230 μmol/L (5). A positive monotonic association between fructosamine and all-cause mortality was previously found when adjusted for albumin (10), similar to the correlation of AlbF with all-cause mortality in diabetic patients or nondiabetic participants in this study. Positive monotonic associations were also observed between fructosamine, AlbF, FPG, and non-CVD mortality in nondiabetic participants. However, associations between fructosamine, AlbF, and FPG with CVD mortality were unremarkable, contrasting with reports of associations between fructosamine and CVD mortality in hemodialysis patients (10) and in less older Chinese adults (40). The results also differ from findings in adults with a mean age lower than 80 years; namely, participants with the lowest quartile of fructosamine and AlbF having higher risks of all-cause and non-CVD mortality (18). Similar results were found in quintile analysis. In contrast to studies of AlbF or fructosamine adjusted for albumin (10,18), significant associations with CVD mortality were not observed in all analyses undertaken here. The potential reasons for these differences may be due to (a) the very elderly in this study tended to have lower levels of FPG than younger Chinese adults; (b) prognostic glycemic factors were different among the general very elderly and dialysis patients; and (c) the very elderly, with limited life expectancy, might be more vulnerable to short-term risks of non-CVD events.

Results of HR values, cumulative mortality, and 5-year mortality revealed that AlbF arranged by quintiles is a better indicator than fructosamine and FPG in the risk classification of all-cause or non-CVD mortality. The performance of AlbF was similar to previous studies investigating fructosamine, glycated albumin, or HbA1c in predicting all-cause mortality among adults with mean age lower than 80 years (4,5). In our study, model performance in predicting mortality was improved after including glycemic markers. Improvements in C-statistics were similar for adding AlbF or FPG to the analysis and better than fructosamine in predicting all-cause or non-CVD mortality among nondiabetic participants. Improvements in IDI and NRI were observed when AlbF was included in the analysis among nondiabetic participants. Performance of AlbF was better than that of fructosamine, which is likely to be affected by variability of serum protein concentrations and fluctuations in albumin (10,41). Our results add to the evidence showing that fructosamine analysis requires correction for serum proteins (8,10,12,41). Significant improvements in IDI and NRI were also observed with AlbF compared to FPG, the latter being the American Diabetes Association-recommended glycemic marker for older adults with coexisting illness and functional limitations (1).

Several guidelines recommended health practitioners to focus on short-term risks and use alternative glycemic markers for HbA1c among diabetic patients with limited life expectancy or multiple coexisting chronic illnesses (1–3). To assess the potential values of AlbF as an alternative glycemic marker in evaluating short-term risk of mortality among the very elderly, we compared the performance of AlbF and FPG among 2 238 very elderly (90.8% were nondiabetic participants) during a median follow-up of 3.0 years (IQR: 1.8–4.3) and found positive monotonic associations between AlbF, FPG, and all-cause or non-CVD mortality in nondiabetic participants. In quintile analysis, the results indicated that abnormal or even high normal ranges of FPG (4.78–5.35 or ≥5.36 mmol/L) were associated with a higher risk of non-CVD mortality in nondiabetic participants. Similarly, nondiabetic participants in the fifth quintiles (≥6.38 μmol/g) of AlbF suffered from significantly higher risk of all-cause and CVD mortality.

Nonlinear analysis also indicated positive monotonic associations between AlbF with all-cause and non-CVD mortality in nondiabetic participants. Moreover, the risks of all-cause and non-CVD mortality were significantly higher in the very elderly with AlbF ≥6.20 and 6.23 μmol/g (Supplementary Figure 4). Similar results were found between AlbF and all-cause mortality in diabetic patients, but notably, due to insufficient sample size, this finding needs to be further validated in a larger population. A similar positive monotonic association was previously revealed between fructosamine (adjusted for albumin) and all-cause mortality in 503 hemodialysis patients aged about 60 years, of whom 216 were nondiabetic participants (10). However, the evidence is currently limited for associations between AlbF and mortality. Both the observational study design and poor correlation between FPG and AlbF in nondiabetic participants make it difficult to define a glucose-lowering target based on AlbF alone. Thus, these findings need to be further validated in future studies.

The strengths of our study include the large sample size of the very elderly, the prospective research design, and abundant covariates on individual characteristics, lifestyles, self-reported diseases, functional health, physical examination, and laboratory tests. These data enabled us to extensively adjust for potential confounders. However, our study also has some limitations. First, cause of death information was only available for 73.2% (872/191) of deaths and conceivably, the missing 26.8% may have resulted in an underestimation of total CVD-related deaths. Consequently, this misinterpretation could distort the actual risk associations between glycemic markers with CVD and non-CVD mortality. Second, the median follow-up time was limited to 3.0 years (IQR: 1.8–4.3), but notably, this time window is appropriate for the very elderly, with limited life expectancy and coexisting illnesses. Third, performance of the oral glucose tolerance test (OGTT), HbA1c, and glycated albumin in predicting mortality was not examined in this study. OGTT is challenging to perform in this population, as multiple blood samples need to be taken within a few hours. Multiple studies suggest that HbA1c is an unstable marker that can be affected by erythrocyte life span, hemoglobin variants, anemia, kidney diseases, liver diseases, iron deficiency, etc. We therefore compared our results of fructosamine and AlbF with FPG as a standard marker of glycemic control. Fourth, although information on the duration of diabetes was not collected in this study, we adjusted for comorbidity status to counteract the potential influence of cumulative hyperglycemic exposure. Fifth, the association between AlbF and mortality might be confounded by the prevalence of liver diseases in the study population.

In conclusion, our results indicated higher glycemic levels among the very elderly were associated with higher all-cause and non-CVD mortality. We found that AlbF rather than FPG was significantly associated with all-cause mortality in the very elderly without diabetes, suggesting that AlbF might be an alternative glycemic marker in predicting mortality among the very elderly without diabetes. Our results provide further supporting evidence for using AlbF in situations where HbA1c may be less accurate.

Funding

This work was supported by the National Natural Science Foundation of China (grant numbers 82025030, 81941023, and 81872707); National Institute on Aging (NIA; grant number 2P01AG031719); United Nations Fund for Population Activities; and Claude D. Pepper Older Americans Independence Centers grant (grant number 5P30 AG028716 from NIA to V.B.K.). The funders played no role in the study design or implementation; data collection, management, analysis, or interpretation; manuscript preparation, review, or approval; or the decision to submit the manuscript for publication.

Conflict of Interest

None declared.

Author Contributions

X.M.S. designed this study. J.H.Z., Y.B.L., F.Z., C.C., F.L., Y.C.L., C.C.L., H.G., Z.X.Y, Z.J.C., C.M., and X.M.S. contributed to the acquisition of the data. J.H.Z., Y.B.L., C.M., and X.M.S. contributed to the analysis and interpretation of the data. J.H.Z. and Y.B.L. wrote the first draft of the manuscript. J.H.Z., Y.B.L., Y.W., X.G., V.B.K., C.M., and X.M.S. critically revised the manuscript for important intellectual content. All authors contributed to multiple drafts and have read and agreed to the final version of this article. X.M.S. takes full responsibility for the work as a whole, including the study design, access to data, and the decision to submit and publish the manuscript.