Biochemical and Hematological Correlates of Elevated Homocysteine in National Surveys and a Longitudinal Study of Urban Adults

Abstract

Elevated blood homocysteine (Hcy) among middle-aged adults can increase age-related disease risk, possibly through other biochemical and hematological markers. We selected markers for hyperhomocysteinemia among middle-aged adults, studied time-dependent Hcy-marker associations and computed highly predictive indices of hyperhomocysteinemia, with cross-sectional and longitudinal validations. We used data from the National Health and Nutrition Examination Survey (NHANES III, phase 2, n_max = 4000), the NHANES 1999-2006 (n_max = 10,151) and pooled NHANES (cross-sectional validation). Longitudinal validation consisted of mixed-effects linear regression models (Hcy predicting markers' annual rates of change), applied to the Healthy Aging in Neighborhoods of Diversity Across the Life Span (HANDLS, n = 227-244 participants, k = 2.4 repeats/participant, Age_base: 30-65 years) data. Machine learning detected nine independent markers for Hcy > 14 µmol/L (NHANES III, phase 2): older age; lower folate and B-12 status; higher serum levels of creatinine, uric acid, alkaline phosphatase, and cotinine; mean cell hemoglobin and red cell distribution widths (RDW); results replicated in the 1999-2006 NHANES [AUC = 0.60-0.80]. Indices combining binary markers increased elevated Hcy odds by 6.9-7.5-fold. In HANDLS, first-visit Hcy predicted annual increase in creatinine, RDW and alkaline phosphatase, with third-visit index (2013-2018) directly predicting Hcy (2004-2009). We provide evidence of the internal and external validity of indices composed of several biomarkers that are strongly associated with elevated Hcy.

Keywords: aging; biochemical indices; hematological indices; homocysteine; inflammation; predictive models.

Figure 1

Flow diagram of predictive modeling using LASSO, ROC curves and multivariable regression modeling. Abbreviations: cvLASSO = cross-validation LASSO; LASSO = least absolute shrinkage and selection operator; LOWESS = Locally weighted regression; HANDLS = Healthy Aging in Neighborhoods of Diversity Across the Life Span; minBIC LASSO = minimum Bayesian information criterion LASSO; NHANES = National Health and Nutrition Examination Surveys; R2 = coefficient of determination.

Figure 2

Nine-marker index (Index I) and its predictive value of elevated Hcy: ROC curves for pooled NHANES ¹. ¹ n = 13,822; optimal cut-point was at 5; AUC = 0.799, 95% CI: 0.785,0.814. Index I included binary biomarkers of elevated Hcy selected using LASSO and backward elimination. The full list of the nine components of Index I are: Age, serum folate, serum vitamin B-12, serum creatinine, red cell distribution width, mean cell hemoglobin, serum cotinine, serum uric acid and alkaline phosphatase. Cut-points for individual components are: serum folate, Log_e, in nmol/L, <2.83; serum creatinine, Log_e, in µmol/L ≥4.481; older age, in years, ≥49; serum vitamin B-12, Log_e, in pmol/L, <5.74; mean cell hemoglobin, Log_e, in pg, ≥3.422; red cell distribution width, Log_e, in %, ≥2.553; serum uric acid, Log_e, in µmol/L, ≥5.826; serum alkaline phosphatase, Log_e, in µmol/L, ≥4.356 U/L; serum cotinine, Log_e, in ng/mL, −0.579.

Figure 3

Eight-marker index (Index II) and its predictive value of elevated Hcy: pooled NHANES ¹. ¹ n = 13,920; optimal cut-point at 5; AUC = 0.798; 95% CI: 0.783,0.812. The index consisted of a summation of all binary Index I biomarkers (See Figure 2 footnotes for cut-points), excluding blood cotinine, which was not available in HANDLS.

Figure 4

Lowess smoother of total Visit 1 homocysteine vs. Visit 3 Index II (r = 0.34, n = 81): HANDLS 2004–2009 (Hcy) and 2013–2018 (Index II) ¹. Abbreviations: HANDLS = Healthy Aging in Neighborhoods of Diversity Across the Life Span: Hcy = Homocysteine; NHANES = National Health and Nutrition Examination Surveys; ROC = Receiver Operating Characteristics analysis. ¹ Blood total Hcy, in µmol/L, is Log_e-transformed and uncategorized. Index II was computed using cut-points obtained from the pooled NHANES ROC analysis. See Table 2 footnotes for details. Index II may range from 0 to 8, and no cut-point was used in this analysis.