Association of Leukocyte Telomere Length With Mortality Among Adult Participants in 3 Longitudinal Studies

Key Points Question Is leukocyte telomere length associated with the natural life span of contemporary humans? Findings This cohort study included 3259 participants from 3 longitudinal studies, of whom 1525 died during the follow-up period. Leukocyte telomere length–associated mortality from noncancer causes increased as participants aged, approaching their age at death. Meaning These data suggest that leukocyte telomere length is associated with a life span limit among contemporary humans.


Introduction
The debate on the natural life span limit in humans has focused on demographic trends 1-5 rather than on biological factors that set a ceiling for life span. We hypothesized that leukocyte telomere length (LTL) might be a biological driver of life span because LTL is associated with increased mortality among older individuals [6][7][8][9][10] and converging evidence infers a causal role of LTL in aging-related diseases that often result in death. 11 The view that LTL plays a causal role in aging-related diseases draws on the following findings.
First, LTL variation across individuals as well as some underlying determinants of LTL variation, including high heritability and sex, are similar in newborns and adults. 16 Second, individuals who enter adult life with short or long LTL are likely to have short or long LTL, respectively, throughout their remaining life course. 17,18 Therefore, having comparatively short or long LTL is principally determined early in life, typically decades before disease onset and mortality. Third, genome-wide association studies have identified LTL-associated single-nucleotide polymorphisms mapped to several regions that harbor telomere maintenance genes. 11,19,20 These single-nucleotide polymorphisms have been used to develop genetic risk scores that show an inverse association of LTL with cardiovascular disease (CVD) [11][12][13]21 and direct associations with some cancers. 12,14,15,21 Such genetic findings largely exclude reverse causality, ie, that CVD might shorten LTL or some cancers might lengthen LTL. Jointly, these findings suggest that LTL is likely causal for CVD and some cancers, perhaps increasing the mortality risk that arises from these diseases. In addition, based on empirical and theoretical considerations, our previous work showed that a subset of the general population may reach a critically short LTL, a so-called telomeric brink, at an age younger than life expectancy, which denotes a high risk of death in the near term. 22

LTL Measurements
Measurements were performed at a baseline examination by Southern blot of the terminal restriction fragments. 26 The interassay coefficients of variation were 2.4%, 1.5%, and 2.0% for the FHS, CHS, and WHI, respectively. In the CHS, 963 individuals had LTL measurements from blood samples obtained in year 5. Among those, 612 (63.6%) had a second measurement in year 10. In the analyses reported in this paper, we used the second LTL measurement for the latter group and the first (and only) LTL measurement for the remaining 351 individuals in the analyzed CHS sample. We also performed sensitivity analyses including only the 612 individuals with 2 LTL measurements. These showed qualitatively similar results; therefore, they are not reported here.

Statistical Analysis
We used t tests for comparisons between the LTL of women and men as well as between the LTL of those who died and those who were alive at the end of follow-up. For the former, we used the ageand study-adjusted LTL computed as follows: we regressed LTL on age and study (ie, as a categorical variable with 3 levels, 1 for each study) and added residuals from this regression to the mean LTL across all individuals in the sample. For the latter, we added the residual LTL (rLTL) to the mean LTL across all individuals in the sample. The rLTL was computed as the residuals from linear regressions of LTL on age, fitted separately among women and men in each of the 3 studies. We also fitted the regressions with quadratic terms for age, but these were nonsignificant in all cases; therefore, we proceeded with the linear model. The same values (with added mean LTL in the entire sample) were used in computations of LTL for individuals who died from cancer and noncancerous causes and those alive at the end of follow-up. We also computed the Pearson correlation coefficient between age at blood draw and sex-and study-adjusted LTL (calculated as the residuals from the regression of LTL on sex and the study variable, added to the mean LTL across all individuals in the sample).
We fitted Cox proportional hazards models using follow-up data on mortality in the combined sample. Time since blood draw was used as the time variable. The most parsimonious model included sex and rLTL as covariates. We used 2 flexible specifications to include age in the model: 1 with a natural spline basis for age and another with age included as a linear term, stratified by baseline age, thus allowing for different baseline hazards in each age strata. Both methods showed similar results for the association of rLTL with mortality. In this article, we report results for the model with splines.
The results for the second approach appear in eTable 3, eTable 5, eTable 10, eFigure 1, eFigure 2, eFigures 5 to 7, and eFigure 9 in the Supplement. For technical details and a description of sensitivity analyses, see the eAppendix in the Supplement.
We analyzed data on cause-specific mortality in the competing risks context using the causespecific hazards functions approach. 27 We used the same model specifications as in the all-cause mortality analyses and estimated respective regression parameters for different cause-specific hazards functions (ie, CVD, OC, and cancer). We report results for the model with splines in the Results section; the results for the second approach appear in eTable 3, eTable 5, eTables 11 to 13, eFigure 1, eFigure 2, eFigure 7, and eFigure 9 in the Supplement. For technical details and description of sensitivity analyses, see the eAppendix in the Supplement.
Statistical analyses with the Cox proportional hazards models and the competing risks models were performed in R version 3.6.1 using the survival package (R Project for Statistical Computing).
Figures were prepared in MATLAB R2019a (MathWorks) and in R version 3.6.1. Statistical significance was set at P < .05, and all tests were 2-tailed.

Results
The   Figure 1A. Participants who survived to the end of follow-up showed a significantly longer median LTL compared with those who died from noncancer causes but compared     with those who died from cancer (alive: 6.69 kb; 95% CI, 6.66-6.72 kb; cancer-related death: 6.61 kb;

JAMA Network Open | Genetics and Genomics
95% CI, 6.55-6.67 kb; noncancer-related death: 6.59 kb; 95% CI, 6.56-6.62 kb) ( Figure 1B). However, we observed minimal association of LTL with cancer mortality ( Figure 2D;  We also performed analyses in separate samples (ie, CHS, FHS, WHI) (eAppendix, eTable 4, and eTable 5 in the Supplement). Figure 4 shows the HRs for a 1-kb decrease in rLTL for each study   Given that different studies had different durations of follow-up (Table), we repeated all calculations, truncating follow-up at 15 years (ie, approximately the mean follow-up in the sample).

JAMA Network Open | Genetics and Genomics
All results were similar to those presented in the text (eFigure 8 and eFigure 9 in the Supplement).
We found evidence of selection among individuals older than 80 years (variances in the groups aged <80 years vs Ն80 years: 0.40 and 0.29; Levene test for equality of variances for the groups: P < .001) (eFigure 10 in the Supplement). This selection is likely owing in part to the earlier deaths among individuals who died from cancer, given that the median age at death from cancer in the analyzed sample was younger than the median age of death from CVD or OCs (81.0 years vs 86.6 years; Kruskal-Wallis test for equality of medians in these groups: P < .001), in line with findings in the general US population. 28,29 Consistent with this premise, rLTLs for individuals who died of cancer at ages younger than 80 years were longer than the rLTLs for those who died of noncancer causes at the same age (6.65 kb vs 6.51 kb; P = .02).   Models in which age was included as a linear term, stratified by baseline age, allowed for different baseline hazards in each age strata. These had similar results as the main analysis (eAppendix, eTables 2-5, eFigure 1, and eFigure 2 in the Supplement).

Discussion
The debate among demographers on the natural life span limit in humans detracts from a more persistent question about biological factors that may determine such a limit. The potential roles of these factors must be considered in the context of specific causes of death. This study showed that short LTL was associated with increased mortality risk as individuals approached the upper boundary of their longevity, a phenomenon principally associated with mortality from noncancer causes.
In absolute terms, the HRs associated with short LTL rapidly escalated as an individual's age Given common misclassifications of cause of death based on death certificates, 30,31 accurate determination of the cause of death was critical for our conclusion that LTL was more strongly associated with death from noncancer causes than death from cancer. That said, death among the older individuals, even when carefully adjudicated, is often not a consequence of a single disease. For instance, stroke or myocardial infarction may occur in different clinical settings among individuals who have multiple health problems (eg, frailty, loss of ambulation due to a fall, diabetes, dementia, infection, etc) that collectively contribute to the individual's death. Regardless of these specific circumstances, it is clear that having comparatively short LTL was associated with increased mortality risk from noncancer causes (ie, CVD and OCs).
Regarding the minimal association of LTL with cancer mortality, we note that, whereas comparatively long LTL [32][33][34] and alleles associated with a long LTL 12,14,15 have been reported to be associated with increased risk of several cancers, short LTL has been reported to be associated with diminished survival among patients with some but not all cancers. [35][36][37][38] Hence, the association of LTL with cancer mortality is complex and contextual; it may reflect opposing telomere-related elements that modify cancer risk, outcome of cancer treatment, and survival.

Limitations
This study has limitations. Our findings are based on individuals of European ancestry who reside in the United States. These results therefore need replication in other groups and geographic locations, given that there is some evidence that the association of LTL with mortality might be influenced by ethnicity. 25 In addition, our analyses did not adjust for key risk factors that are associated with mortality risk (eg, hypertension, dyslipidemia, diabetes, smoking, obesity) or for comorbidities that may have been present at baseline and also contributed to mortality (eg, CVD and cancer).