The Longer, the Better? An Empirical Study of the Extent and Mechanisms of Attenuating Biomarker Associations in Cardiovascular Patient Cohorts.
The search for novel prognostic markers and potential drug targets in chronic disorders such as coronary heart disease (CHD) [7] is a highly active research field. Typically, candidate markers implicated in the condition of interest based on more or less elaborate pathophysiological knowledge are carried forward to observational cohort studies to examine their prognostic potential and/or etiological role, as soon as assessment in a substantive number of samples becomes feasible (1-3). Timely completion is of utmost importance owing to possibly substantial clinical implications, and the apparent advantages of conducting additional long-term follow-up leading to more robust results and higher statistical precision need to be balanced against this well-justified urgency.Despite the intuitive appeal of longer follow-up yielding more outcome events, longer follow-up may also mean that the association of interest becomes attenuated. This phenomenon, related mainly to time-variability of exposures as well as their effects, may have important implications for the design and interpretation of longerterm follow-up studies (4). The impact on the interpretation of etiologic studies could be particularly severe, if the actual reasons for the rapid attenuation of estimated hazard ratios over increasing follow-up were reverse causality and/or differential survival (5). In practice, however, once the association of a novel biomarker with prognosis has been quantified in a certain study, the actual impact of completing additional follow-up is hardly ever explored or even systematically analyzed. This might compromise patient cohort studies of novel markers in particular, for which a certain degree of ignorance of temporal marker dynamics following baseline and/or foregoing overt outcome events is the norm. Even so, studies addressing these issues in general population cohorts are scarce (6-8), and there is an absolute dearth of pertinent research in the context of patient cohort settings, possibly because comprehensive empirical illustrations of this issue are lacking.
In the Langzeiterfolge der KARdiOLogischen Anschlussheilbehandlung, (KAROLA; Long-term success of cardiological rehabilitation therapy) prospective cohort study of patients with CHD, numerous novel and established prognostic biomarkers have been addressed since its beginning. The first longitudinal analyses were based on 3 years of follow-up (9), but follow-up is ongoing and published analyses encompass up to 10 years of follow-up (10). In the present work, we reanalyze the prognostic associations of a variety of lipids, metabolic, inflammatory, as well as cardiac and other biomarkers with prognosis after up to 13 years, with special focus on the impact of increasing follow-up duration on the empirical estimates of association. The roles of reverse causality and differential survival are explored, as well as the implications for statistical power over increasing follow-up durations.
Methods
STUDY DESIGN AND DATA COLLECTION
Recruitment for the KAROLA prospective cohort study was done from January 1999 to May 2000, and included patients aged 30-70 years who were admitted to 1 of 2 participating rehabilitation clinics for inpatient rehabilitation within 3 months after an acute cardiovascular event [acute coronary syndrome (ACS) or coronary artery bypass grafting (CABG)]. All participants gave written informed consent. Ethical approval was obtained from the ethics boards of the physicians' chambers of Hessen and Baden-Wurttemberg, the University of Ulm, and the University of Heidelberg. The study protocol complied with the Declaration of Helsinki.
Information on data collection, measurements, and distribution of biomarkers is provided in the Supplemental Methods that accompany the online version of this article at https://www.clinchem.org/content/vol63/issue3. Questionnaire follow-up was done after 1, 3, 4.5, 6, 8, 10, and 13 years, whereas repeat blood samples were obtained after 1 and 3 years.
STATISTICAL ANALYSIS
All KAROLA participants for whom mortality follow-up was available were included in the present analysis. The outcome analyzed in the present study was general mortality, which was visualized using Kaplan--Meier plots of the time from baseline to death from any cause. For survivors, observation time was censored at the last date that a subject was known to be alive, combining information from patients, physicians, and public health authorities.
Adjusted associations of the various markers with mortality were quantified by multiple Cox regression. As the purpose of the present study was not to judge the prognostic value or causal role of any individual biomarker per se, but rather to empirically examine the impact of the length of follow-up on the observed associations, exposure coding and covariable adjustment generally followed the main analyses in the primary study reports. Thus, if the original paper presented the age-, sex-, and BMI-adjusted association of the candidate marker categorized into quartiles as the main model, the present study likewise used this adjustment set and marker quartiles. For markers that were analyzed as continuous variables, estimates were presented per SD to increase comparability and readability of the results.
The Cox models were first fit using all follow-up data available, subsequently truncating the observation time at 3, 4.5, 6, 8, or 10 years. The different hazard ratio estimates were then tabulated by follow-up/truncation time (for the sake of clarity, we focused on the estimates pertaining to the most relevant coding of each marker, reporting, for example, only the hazard ratio for the third, in reference to the first marker tertile, or the hazard ratio per SD, etc.). To summarize the changes of the estimated hazard ratio with increasing follow-up duration, a trend of attenuation measure y was estimated for each marker by fitting a simple nonlinear model of the form [[beta].sub.t] = [[gamma].sub.t] X [[beta].sub.0], where t indicates the analyzed follow-up duration in years and [[beta].sub.t] is the corresponding empirical log(hazard ratio) estimated by Cox regression. Some properties of [gamma] are illustrated in online Supplemental Fig. S1. In brief, [gamma] = 1 indicates no attenuation, and attenuation occurs more rapidly the closer [gamma] comes to 0. To explore the interplay of attenuating hazard ratios and statistical power, the latter was calculated based on the formula provided by Collett (11) for a reasonable set of scenarios, in which a binary risk factor with hazard ratio of 2.5, 2.0, or 1.75 after 2 years of follow-up was assumed to be subject to varying degrees of attenuation.
Differential attrition was explored by tabulating the frequency and proportion (for selected continuous markers: mean and CI) of the various risk indicator categories for the overall study population, as well as in the subpopulations that were still included in the study after 3, 6, or 10 years of follow-up. For the sake of clarity, these analyses were limited to disease severity indicators (number of affected vessels, left ventricular function impairment) and biomarkers that had shown particularly pronounced attenuation in the abovementioned analyses. To approach the topic of time-varying associations and reverse causality, age- and sex-adjusted Cox regression models predicting mortality from baseline characteristics were then fit to non-overlapping follow-up periods of 3 to 4 years' duration (years 1-3; years 4-6; years 7-10; years 11-13). For markers that had been measured repeatedly, the association of baseline marker values with mortality during years 1-3 or years 4-6 was compared to the association of marker values measured after 3 years of follow-up with mortality during years 4-6. Subsequently, extended Cox models allowing for time-varying predictor values were used to examine the impact of incorporating marker measurements obtained after 1 and 3 years in models of mortality during overall follow-up.
Observations with missing values were excluded only from analyses in which the respective variable was required. CIs were based on an [alpha]-level of 0.05. SAS 9.3 was used for Cox modeling and descriptive statistics. Nonlinear estimation was done with R 3.0.2.
Results
STUDY POPULATION AND SURVIVAL
A total of 1204 KAROLA participants (16% female, median age 61 years) had mortality follow-up available and were included in the present study [for comprehensive descriptives, see (9)]. The median time that had passed between the initial cardiovascular event and study inclusion was 41 [interquartile range (IQR) 34-49] days. The median follow-up duration was 13.0 (IQR 12.1-13.1) years, and 276 participants (22.9%) died. As can be seen in online Supplemental Fig. S2, the survival experience was rather similar in male and female participants.
EMPIRICAL IMPACT OF FOLLOW-UP DURATION ON ESTIMATED HAZARD RATIOS
The hazard ratios obtained after 13 years of follow-up, and estimates resulting from truncation of the observation times at the end of previous follow-up rounds, are shown in Table 1 for the adjusted analysis of all-cause mortality. For the majority of markers, there was a clear trend of weakening associations with increasing follow-up duration. Some of these patterns were very pronounced, and the trends overall were rather consistent despite sometimes wide CIs. For 3 markers [parathyroid hormone (PTH), mid-regional pro-atrial natriuretic peptide (MR-proANP), and [gamma]-glutamyltransferase (GGT)], no attenuation was observed, with [gamma]-values very close to 1.00. For asymmetric dimethylarginine (ADMA), free fatty acids (FFAs), and phosphate, hazard ratio point estimates changed directions across the follow-up. Our attenuation model appears ill-suited for this situation, and the nonlinear y estimation often converged on either [gamma] [much less than] 1 (with [[beta].sub.0] [much less than] 0) or [gamma] > 1 (with very small ([[beta].sub.0] > 0) depending on starting values, without clear guidance for interpretation being given by the pattern of empirical hazard ratios. To be conservative, the trend of attenuation y was considered undefined in these cases. As shown in Fig. 1, moderate attenuation of hazard ratios with increasing follow-up ([gamma] = 0.9) regularly results in nonmonotonic relationships between follow-up duration and power to detect an association with some candidate marker. Furthermore, with somewhat more pronounced but still realistic attenuation ([gamma] = 0.8), longer follow-up cannot overcome its adverse effects on statistical power in this simple example scenario resembling our cohort in sample size and incidence.
DIFFERENTIAL ATTRITION BY RISK STRATA
There was consistent evidence of differential attrition by disease severity indicators as well as several biomarkers. For instance, in 3 out of 4 disease severity indicators [left ventricular function, N-terminal pro-B-natriuretic peptide (NT-proBNP), troponin T], [greater than or equal to] 70% of the participants in the highest risk stratum remained in the analysis set for the last 3-year period of follow-up, compared to >80% in the low risk strata (Table 2). Among other biomarkers, differences in the attrition of high- and lowrisk categories were >10% points for high-sensitivity C-reactive protein (hsCRP), type II secretory phospholipase A2 (sPLA2-IIa), and cystatin C (Table 3).
HAZARD RATIOS DURING VARIOUS FOLLOW-UP PERIODS
The hazard ratios for the association of baseline markers with mortality during various periods of follow-up are shown in Table 4. The associations were generally much more pronounced when restricting the analysis to the first 3 years of follow-up. Though generally overlapping CIs must be acknowledged as a reminder against overinterpretation, some distinct patterns with respect to the evolution of hazard ratios during later follow-up periods seemed to emerge. Some markers showed evidence of a substantial association during early follow-up with subsequent absence of any association (for example, number of affected vessels and HDL cholesterol). The other extreme was a strong association during early follow-up with a more or less monotone decrease in hazard ratios over subsequent follow-up periods and with some suggestion of a persistently increased risk (for example, troponin T and cystatin C). Most markers showed less clear patterns, but markers with pronounced differential attrition (Table 2, Table 3) more frequently appeared to yield substantial hazard ratios even during later follow-up.
REPEATED MEASUREMENTS
The impact of considering repeated measurements of markers during follow-up is demonstrated in Table 5. Especially for NT-proBNP, troponin T, and hsCRP, the hazard ratio during follow-up years 4-6 increased substantively when based on measurements taken after 3 years of follow-up rather than at baseline. When the marker measurements taken after 3 years offollow-up, or additionally those taken after 1 year of follow-up, were incorporated in the analysis by fitting extended Cox regression models with time-varying covariables, the hazard ratio estimates pertaining to the full follow-up period generally increased. Correlations between the repeat measurements are shown in online Supplemental Table S1.
Discussion
Our findings demonstrate how associations of candidate biomarkers with clinical outcomes in a cardiovascular patient cohort exhibit substantial attenuation over increasing lengths of follow-up. Although the presence of some such patterns is not altogether surprising, the extent of this issue had not been empirically addressed in the context of clinical cohort studies. Our analyses show that the magnitude of attenuation sometimes is rather extreme in such a setting, with important implications for the design, analysis, and interpretation of biomarker studies in cardiovascular and other patient cohorts. Furthermore, they demonstrate how the role of differential survival and reverse causality may be tentatively approached through explicit consideration of varying lengths and periods of follow-up.
Differential attrition appeared rather pronounced in several biomarkers and was also common in disease severity indicators in the present study. If differential attrition occurs, for example, because high risk participants are clustered in some (noncausal) marker category at baseline and are removed preferentially from the study because of their high-risk profile, the respective marker category will successively be depleted of high-risk participants and become more and more similar to the reference category, producing an attenuation of hazard ratios over increasing follow-up. Such differential attrition, however, appears an insufficient explanation for the attenuation of hazard ratios over longer follow-up in the present study. In particular, the markers with relevant hazard ratios persisting throughout follow-up tended to show more, rather than less, pronounced differential attrition. Differential attrition due to excess outcomes or mortality indeed is an important characteristic of genuine risk markers, especially if the markers are stable in the long-term, and if single baseline measurements function well by locating a person on a pathophysiologically relevant trajectory not overly affected by subsequent disturbances.
The availability of repeated measurements over follow-up may allow for disentangling of some of the mechanisms leading to attenuated associations. For example, some of the hazard ratios during the second 3-year period of follow-up resembled those during the first 3 years in the present study, if the analysis was based on repeat measurements obtained after 3 years of follow-up instead of at baseline, which suggests consistent shortterm predictiveness without relevant distortion by the baseline acute event. A pattern with stronger marker outcome associations during early follow-up, and weaker predictiveness of later marker measurements instead could lead to the consideration of reverse causality, though, it could also indicate true short-term predictiveness limited to the time period immediately following the baseline acute event. Bearing in mind the overlapping of CIs and the need for independent replications, the distinct patterns observed allow for drawing tentative further conclusions. For instance, whereas the hazard ratio of baseline hsCRP essentially vanished for the second 3-year period, this was not so much the case for troponin T, which could be interpreted as baseline hsCRP carrying only much shorter-term information than troponin T. The much stronger hazard ratio of troponin T during the first 3 years might suggest troponin T to be subject to reverse causality, which would appear as less of an issue for hsCRP or cystatin C, for which the hazard ratios during the second 3 years resembled the early follow-up associations if based on the measurements after 3 years. Finally, the intriguingly stronger association of NTproBNP during later follow-up could result from baseline and follow-up measurements reflecting somewhat different phenomena (post event acute impairment vs long-term cardiac remodeling).
The extended Cox proportional hazards model can readily accommodate repeated measurements during follow-up [e.g., (12)]. Though common and certainly correct in many situations, this approach is based on the assumption that the association of a risk marker and outcome is constant over the entire follow-up period. The marker-risk relationship may often be more complicated in patient cohorts, where, for example, high baseline hsCRP might tend to indicate disease severity (and, thus, cardiovascular risk), and some high follow-up hsCRP might tend to indicate unspecific environmental stimuli [and, thus, not cardiovascular risk (4)]. Such situations will be particularly frequent in secondary event studies, where the primary event may be associated with pronounced temporary alterations of the participants' physiology. For example, beyond the stronger association suggested for NT-proBNP during late follow-up as described in the preceding paragraph, it has recently been shown in the KAROLA study that an increase of NTproBNP during the first year of follow-up is associated with subsequent cardiovascular risk, independently from baseline NT-proBNP (10). Independent of any primary event, markers such as hsCRP might consistently have short-term predictiveness by reflecting a physiologically relevant state at any individual time point, yet would not necessarily have similarly strong predictiveness in the long-term due to a lack of long-term stability, and because there is not necessarily a correlation of who is in that particular relevant state at different time points.
The above considerations make it clear that an explicit and considerate comparison of hazard ratios estimated for different periods of follow-up, and pertaining to measurements taken at different time points, may yield substantial insights and diverse conclusions in particular in patient cohort studies. Extended Cox regression modeling with time-varying predictors may have more limitations than strengths in this context (12, 13). The average estimate obtained in such a model may be rather misleading, regardless of the main focus and intended application of the study: investigations of the predictive utility of a marker will most commonly be aimed at a setting with a single baseline measurement (where stronger associations brought about by repeated measurements during later follow-up would be misleadingly overoptimistic), whereas investigations on causal relationships need to consciously address the mutually nonexclusive possibilities of truly time-varying effects and reverse causality. Extended Cox models allowing for nonproportional hazards through time-varying effect estimates, on the other hand, might complicate communicability of results, require difficult decisions on the functional form of the deviation from time-constant associations, and, overall, come with their own caveats and limitations (14). As a side note, testing the proportional hazards assumption by analyzing interactions of the various biomarkers with log(observation time) in the present study yielded a statistically significant result only in the case of triglycerides. This common approach to assessment of the Cox model assumption is one example of allowing for a specific deviation from proportional hazards, but it fails to uncover the intriguing patterns discussed in the present work.
Being a topic somewhat remotely related to the issues studied in the present work, regression dilution refers to the systematic underestimation of regression coefficient due to (unmeasured) within-person variability of biomarkers (15). As the extent of regression dilutions seems to be rather consistent across studies [see, for example, (6, 7, 16-18)], it has been suggested to use published estimates of regression dilution to correct the analyses in other cohorts in which repeated measurements are not available (19), but such data are available less frequently if novel markers and special cohorts are of interest. Methods correcting for regression dilution based on short-term repeat measurements also are not appropriate or sufficient if the association of interest is not simply due to the average marker value, but if the analysis instead needs to account for time-varying marker levels or time-dependent effects. It also appears dubious whether repeat measurements obtained after substantial follow-up durations can yield useful information on nonbiological within-person variability, which would be the key mea sure of interest in the context of correcting for regression dilution. This would require the assumption of an extreme long-term stability of biomarker values, and once again one might be even less willing to make such stability assumptions in the setting of patient cohorts defined by a severe disease event at baseline.
The present investigation was limited to selected markers of cardiovascular risk. However, these encompassed a variety of disease severity indicators as well as established and novel biomarkers of cardiovascular risk. CIs often were overlapping, and some patterns might become clearer if future studies could be conducted in larger patient cohorts and with a more comprehensive set of biomarkers, potentially better capturing the underlying pathophysiology (20). Furthermore, only a limited number of repeated measurements were available. Future pertinent work could benefit from a larger number of both immediate and long-term repeated measurements, which would allow a comprehensive dissection of the various sources of measurement variability, as well as developing a more refined attenuation model. Our covariable adjustment generally followed the reports originally analyzing the biomarkers of interest, which ignored medication that could potentially change during follow-up and could also play a role in the attenuation of effects. Finally, one should avoid generalizing the results from our secondary prevention setting to generally healthy populations and primary prevention. Despite these limitations, the present study demonstrated in quite some detail the extent and complexity of attenuating associations over follow-up in long-term patient cohorts, an issue that remains by and large ignored in all but the rarest pertinent studies.
In conclusion, the attenuation of estimates of association in prognostic and etiologic cardiovascular patient cohort studies appears to be of much greater relevance than currently appreciated. Prognostic studies should include at least some repeat measurement. If long-term prediction is of special interest, repeated baseline assessments reducing measurement error should be a reasonable priority. If causality issues are to be addressed, a conscious spacing of repeat measurements relative to the initial event might be particularly important, but, again, the analysis must not be focused on an averaged hazard ratio obtained by extended Cox regression. Altogether, unless long-term association are of primary interest, e.g., in life course or long-term medication studies, a larger sample size might seem more advantageous than a longer follow-up in more situations than previously appreciated. Future work is needed to clarify how these competing aspects of study design can best be balanced against one another in the light of the pronounced attenuation of associations that may occur especially in the context of cardiovascular patient cohorts.
Author Contributions: All authors confirmed they have contributed to the intellectual content of this paper and have met the following 3 requirements: (a) significant contributions to the conception and design, acquisition of data, or analysis and interpretation of data; (b) drafting or revising the article for intellectual content; and (c) final approval ofthe published article.
Authors' Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the author disclosure form. Disclosures and/or potential conflicts ofinterest:
Employment or Leadership: None declared.
Consultant or Advisory Role: W. Koenig, Novartis, Pfizer, The Medicines Company, Amgen, AstraZeneca, MSD, and GSK. Stock Ownership: None declared.
Honoraria: W. Koenig, AstraZeneca, Novartis, MSD, Amgen, Sanofi, Actavis, and Berlin-Chemie.
Research Funding: H. Brenner, German Ministry of Education and Research (01GD9820/0 and 01ER0814) and Willy-Robert-Pitzer Foundation to the institution.
Expert Testimony: None declared.
Patents: None declared.
Role of Sponsor: The funding organizations played no role in the design of study, choice of enrolled patients, review and interpretation of data, and final approval of manuscript.
References
(1.) Apple FS, Wu AHB, Mair J, Ravkilde J, Panteghini M, Tate J, etal. Future biomarkersfor detection of ischemia and risk stratification in acute coronary syndrome. Clin Chem 2005;51:810-24.
(2.) May A, Wang TJ. Biomarkersfor cardiovascular disease: challenges and future directions. Trends Mol Med 2008;14:261-7.
(3.) O'DonoghueM, Morrow DA.Thefuture of biomarkersin the management of patients with acute coronary syndromes. Curr Opin Cardiol 2008;23:309-14.
(4.) Koenig W, Sund M, Frohlich M, Lowel H, Hutchinson WL, Pepys MB. Refinement of the association of serum C-reactive protein concentration and coronary heart disease risk by correction for within-subject variation over time: the MONICA Augsburg studies, 1984 and 1987. Am J Epidemiol 2003;158:357-64.
(5.) de Vries CL, Feskens EJ, de Lezenne Coulander C, Kromhout D. Repeated measurements of serum cholesterol and blood pressure in relation to long-term incidence of myocardial infarction. The Zutphen Study. Cardiology 1993;82:89-99.
(6.) Clarke R, Shipley M, Lewington S, Youngman L, Collins R, Marmot M, Peto R. Underestimation of risk associations due to regression dilution in long-term follow-up of prospective studies. Am J Epidemiol 1999;150: 341-53.
(7.) The Fibrinogen Studies Collaboration, Wood AM, White I, Thompson SG, Lewington S, Danesh J. Regression dilution methodsfor meta-analysis: assessing long-term variability in plasma fibrinogen among 27247 adults in 15 prospective studies. Int J Epidemiol 2006;35:1570-8.
(8.) MacMahon S, Peta R, Cutler J, Collins R, Sorlie P, Neaton J, etal. Blood pressure, stroke, and coronary heart disease. Part 1, prolonged differences in blood pressure: prospective observational studies corrected for the regression dilution bias. Lancet 1990;335:765-74.
(9.) Koenig W, Twardella D, Brenner H, Rothenbacher D. Plasma concentrations of cystatin C in patients with coronary heart disease and risk for secondary cardiovascular events: more than simply a marker of glomerular filtration rate. Clin Chem 2005;51:321-7.
(10.) Dallmeier D, Pencina MJ, Rajman I, Koenig W, Rothenbacher D, Brenner H. Serial measurements of N-terminal pro-brain natriuretic peptide in patients with coronary heart disease. PLoS One 2015;10:e0117143.
(11.) Collett D. Modelling survival data in medical research. New York: Chapman & Hall; 1994.
(12.) O'Doherty MG, Jorgensen T, Borglykke A, Brenner H, Schottker B, Wilsgaard T, et al. Repeated measures of body mass index and C-reactive protein in relation to all-cause mortality and cardiovascular disease: results from the Consortium on Health and Ageing Network of Cohorts in Europe and the United States (CHANCES). Eur J Epidemiol 2014;29:887-97.
(13.) Hosmer DW Jr, Lemeshow S. Applied survival analysis: regression modelling of time to event data. New York: Wiley; 1999.
(14.) Therneau TM, Grambsch PM. Modeling survival data: extending the Cox model. New York: Springer; 2000.
(15.) Frost C, Thompson SG. Correcting for regression dilution bias: comparison of methodsfor a single predictor variable. J R Stat Soc Series A 2000;163:173-89.
(16.) Clarke R, Lewington S, Donald A, Johnston C, Refsum H, Stratton I, etal. Underestimation of the importance of homocysteine as a risk factor for cardiovascular disease in epidemiological studies. J Cardiovasc Risk 2001;8:363-9.
(17.) Law MR, Wald NJ, Wu T, HackshawA, Bailey A. Systematic underestimation of association between serum cholesterol concentration and ischaemic heart disease in observational studies: data from the BUPA study. BMJ 1994;308:363-6.
(18.) Lewington S, Thomsen T, Davidsen M, Sherliker P, Clarke R. Regression dilution bias in blood total and high-density lipoprotein cholesterol and blood pressure in the Glostrup and Framingham prospective studies. J Cardiovasc Risk 2003;10:143-8.
(19.) Emberson JR, Whincup PH, Morris RW, Walker M, Lowe GDO, Rumley A. Extent of regression dilution for established and novel coronary risk factors: results from the British Regional Heart Study. Eur J Cardiovasc Prev Rehabil 2004;11:125-34.
(20.) Duprez DA, Otvos J, Sanchez OA, Mackey RH, Tracy R, Jacobs DR Jr. Comparison of the predictive value of Glyc A and other biomarkers of inflammation for total death, incident cardiovascular events, noncardiovascular and noncancer inflammatory-related events, and total cancer events. Clin Chem 2016;62:102031.
Caption: Fig. 1. Power to detect the association with a binary risk factor (affecting half the population) after varying length of follow-up, assuming n = 1200 and an incidence proportion of 3% per year, for 3 different initial hazard ratios (HR); black: association not affected by attenuation; blue: HR attenuating with [gamma] = 0.9; orange: HR attenuating with [gamma = 0.8.
Lutz P. Breitling, [1] * Ute Mons, [1] Harry Hahmann, [2] Wolfgang Koenig, [3,4,5] Dietrich Rothenbacher, [6] and Hermann Brenner [1]
[1] German Cancer Research Center, Division of Clinical Epidemiology and Aging Research, Heidelberg, Germany; [2] Schwabenland-Klinik, Isny-Neutrauchburg, Germany; [3] University of Ulm Medical Center, Department of Internal Medicine II - Cardiology, Ulm, Ger many; [4] Deutsches Herzzentrum Munchen, Technische Universitat Munchen, Munich, Germany; [5] DZHK (German Centre for Cardiovascular Research), partner site Munich Heart Alliance, Munich, Germany; [6] Ulm University, Institute of Epidemiology and Medical Biometry, Ulm, Germany.
[7] Nonstandard abbreviations: CHD, coronary heart disease; KAROLA, Langzeiterfolge der KARdiOLogischen Anschlussheilbehandlung, Long-termsuccess of cardiological rehabilitation therapy; ACS, acute coronary syndrome; CABG, coronary artery bypass grafting; BMI, body mass index; IQR, interquartile range; PTH, parathyroid hormone; MR-proANP, mid-regional pro-atrial natriuretic peptide; GGT, y-glutamyltransferase; ADMA, asymmetric dimethylarginine; FFA, free fatty acid; NT-proBNP, N-terminal pro-B-natriuretic peptide; hsCRP, high-sensitivity C-reactive protein; type II secretory phospholipase A2 (sPLA2-IIa).
* Address correspondence to this author at: German Cancer Research Center, Division C070, POB 101949, D-69009 Heidelberg, Germany. Fax0049-06221-421302; e-mail [email protected].
Received July 6, 2016; accepted August 18, 2016.
Previously published online at DOI: 10.1373/clinchem.2016.263202
Table 1. General mortality adjusted hazard ratios (HRs) estimated by
Cox regression by using the entire follow-up (13 years) or by
truncating at the end of previous follow-up rounds. (a)
3 years
Marker Contrast HR (95% CI)
CETP(b) Quartile 1 vs 4 3.78 (0.96-14.9)
HDL cholesterol Per SD 0.56 (0.37-0.83)
LDL cholesterol Per SD 1.41 (0.99-2.01)
sPLA2-lla mass Per SD (log) 1.52 (1.03-2.24)
Lp-PLA2 mass Tertile 3 vs 1 2.87 (1.17-7.08)
Uric acid Quartile 4 vs 1 3.89 (1.05-14.4)
CRP Per SD (log) 2.05 (1.38-3.03)
Cystatin C Quintile 5 vs 1 4.95 (1.05-23.4)
Triglycerides Per SD 1.50 (1.20-1.88)
SDMA Per SD 1.38 (1.03-1.86)
Cardiac troponin T Quartile 4 vs 1 6.23 (1.34-28.9)
A-FABP Per SD (log) 1.40 (0.89-2.21)
Creatinine clearance <60 vs 90+ mL/min 3.06 (0.78-12.0)
NT-proBNP Quartile 4 vs 1 4.61 (1.26-16.9)
F2RL3 [c] methylation Per SD 1.41 (0.93-2.14)
Calcium Per SD 1.31 (0.80-2.13)
PTH Per SD 1.15 (0.89-1.47)
MR-proANP Per SD (log) 1.46 (0.87-2.45)
GGT Quartile (d) 4 vs 1 1.34 (0.42-4.22)
ADMA Per SD 0.92 (0.62-1.37)
FFA Quartile 4 vs 1 0.88 (0.29-2.60)
Phosphate Per SD 0.70 (0.40-1.25)
4.5 years
Marker Contrast HR (95% CI)
CETP (b) Quartile 1 vs 4 1.68 (0.70-3.99)
HDL cholesterol Per SD 0.77 (0.56-1.05)
LDL cholesterol Per SD 1.28 (0.96-1.72)
sPLA2-lla mass Per SD (log) 1.22 (0.91-1.64)
Lp-PLA2 mass Tertile 3 vs 1 1.88 (0.92-3.85)
Uric acid Quartile 4 vs 1 2.15 (0.93-5.00)
CRP Per SD (log) 1.77 (1.30-2.40)
Cystatin C Quintile 5 vs 1 2.74 (1.05-7.14)
Triglycerides Per SD 1.45 (1.20-1.75)
SDMA Per SD 1.49 (1.22-1.82)
Cardiac troponin T Quartile 4 vs 1 5.79 (1.66-20.2)
A-FABP Per SD (log) 1.35 (0.94-1.94)
Creatinine clearance <60 vs 90+ mL/min 5.00 (1.92-13.0)
NT-proBNP Quartile 4 vs 1 2.63 (1.13-6.15)
F2RL3 [c] methylation Per SD 1.27 (0.91-1.77)
Calcium Per SD 1.29 (0.90-1.85)
PTH Per SD 1.19 (0.98-1.45)
MR-proANP Per SD (log) 1.81 (1.20-2.73)
GGT Quartile (d) 4 vs 1 2.58 (1.07-6.22)
ADMA Per SD 0.99 (0.74-1.33)
FFA Quartile 4 vs 1 1.01 (0.43-2.35)
Phosphate Per SD 1.01 (0.69-1.50)
6 years
Marker Contrast HR (95% CI)
CETP (b) Quartile 1 vs 4 1.41 (0.74-2.69)
HDL cholesterol Per SD 0.81 (0.63-1.03)
LDL cholesterol Per SD 1.05 (0.83-1.34)
sPLA2-lla mass Per SD (log) 1.00 (0.79-1.27)
Lp-PLA2 mass Tertile 3 vs 1 1.43 (0.81-2.51)
Uric acid Quartile 4 vs 1 1.67 (0.86-3.26)
CRP Per SD (log) 1.39 (1.09-1.76)
Cystatin C Quintile 5 vs 1 2.50 (1.19-5.27)
Triglycerides Per SD 1.37 (1.17-1.61)
SDMA Per SD 1.37 (1.16-1.63)
Cardiac troponin T Quartile 4 vs 1 4.54 (1.70-12.1)
A-FABP Per SD (log) 1.40 (1.06-1.85)
Creatinine clearance <60 vs 90+ mL/min 4.00 (1.76-9.11)
NT-proBNP Quartile 4 vs 1 3.33 (1.56-7.12)
F2RL3 [C] methylation Per SD 1.26 (0.97-1.64)
Calcium Per SD 1.41 (1.06-1.86)
PTH Per SD 1.20 (1.02-1.41)
MR-proANP Per SD (log) 1.60 (1.15-2.23)
GGT Quartile (d) 4 vs 1 1.92 (1.04-3.54)
ADMA Per SD 1.06 (0.84-1.32)
FFA Quartile 4 vs 1 0.91 (0.47-1.76)
Phosphate Per SD 1.00 (0.74-1.34)
8 years
Marker Contrast HR (95% CI)
CETP (b) Quartile 1 vs 4 1.25 (0.74-2.14)
HDL cholesterol Per SD 0.92 (0.76-1.11)
LDL cholesterol Per SD 1.10 (0.90-1.34)
sPLA2-lla mass Per SD (log) 1.19 (0.98-1.44)
Lp-PLA2 mass Tertile 3 vs 1 1.41 (0.87-2.27)
Uric acid Quartile 4 vs 1 1.75 (1.03-2.95)
CRP Per SD (log) 1.54 (1.26-1.88)
Cystatin C Quintile 5 vs 1 2.47 (1.36-4.48)
Triglycerides Per SD 1.28 (1.10-1.49)
SDMA Per SD 1.34 (1.15-1.56)
Cardiac troponin T Quartile 4 vs 1 3.84 (1.89-7.83)
A-FABP Per SD (log) 1.48 (1.18-1.85)
Creatinine clearance <60 vs 90+ mL/min 2.97 1.50-5.91)
NT-proBNP Quartile 4 vs 1 3.40 (1.81-6.40)
F2RL3 [C] methylation Per SD 1.29 (1.03-1.60)
Calcium Per SD 1.32 (1.05-1.67)
PTH Per SD 1.20 (1.04-1.38)
MR-proANP Per SD (log) 1.66 (1.26-2.17)
GGT Quartile (d) 4 vs 1 1.96 (1.17-3.28)
ADMA Per SD 1.12 (0.94-1.35)
FFA Quartile 4 vs 1 1.16 (0.67-2.03)
Phosphate Per SD 1.04 (0.83-1.32)
10 years
Marker Contrast HR (95% CI)
CETP (b) Quartile 1 vs 4 1.21 (0.77-1.90)
HDL cholesterol Per SD 0.95 (0.81-1.11)
LDL cholesterol Per SD 1.06 (0.90-1.25)
sPLA2-lla mass Per SD (log) 1.14 (0.97-1.35)
Lp-PLA2 mass Tertile 3 vs 1 1.34 (0.88-2.04)
Uric acid Quartile 4 vs 1 1.57 (1.01-2.45)
CRP Per SD (log) 1.41 (1.19-1.67)
Cystatin C Quintile 5 vs 1 2.32 (1.41-3.83)
Triglycerides Per SD 1.28 (1.13-1.46)
SDMA Per SD 1.23 (1.06-1.42)
Cardiac troponin T Quartile 4 vs 1 3.33 (1.88-5.90)
A-FABP Per SD (log) 1.35 (1.11-1.64)
Creatinine clearance <60 vs 90+ mL/min 2.69 (1.46-4.95)
NT-proBNP Quartile 4 vs 1 3.11 (1.83-5.27)
F2RL3 [C] methylation Per SD 1.33 (1.10-1.61)
Calcium Per SD 1.29 (1.05-1.58)
PTH Per SD 1.15 (1.00-1.31)
MR-proANP Per SD (log) 1.71 (1.35-2.15)
GGT Quartile (d) 4 vs 1 2.05 (1.30-3.22)
ADMA Per SD 1.10 (0.94-1.28)
FFA Quartile 4 vs 1 1.45 (0.88-2.39)
Phosphate Per SD 1.05 (0.87-1.26)
13 years
Marker Contrast HR (95% CI)
CETP (b) Quartile 1 vs 4 1.13 (0.78-1.64)
HDL cholesterol Per SD 0.93 (0.81-1.06)
LDL cholesterol Per SD 1.07 (0.94-1.22)
sPLA2-lla mass Per SD (log) 1.18 (1.03-1.34)
Lp-PLA2 mass Tertile 3 vs 1 1.36 (0.97-1.90)
Uric acid Quartile 4 vs 1 1.81 (1.26-2.60)
CRP Per SD (log) 1.35 (1.18-1.54)
Cystatin C Quintile 5 vs 1 1.98 (1.35-2.88)
Triglycerides Per SD 1.22 (1.10-1.36)
SDMA Per SD 1.20 (1.07-1.34)
Cardiac troponin T Quartile 4 vs 1 2.89 (1.88-4.43)
A-FABP Per SD (log) 1.20 (1.02-1.40)
Creatinine clearance <60 vs 90+ mL/min 2.76 (1.73-4.43)
NT-proBNP Quartile 4 vs 1 2.80 (1.86-4.23)
F2RL3 [C] methylation Per SD 1.24 (1.07-1.44)
Calcium Per SD 1.24 (1.05-1.47)
PTH Per SD 1.19 (1.07-1.32)
MR-proANP Per SD (log) 1.57 (1.30-1.89)
GGT Quartile (d) 4 vs 1 1.73 (1.21-2.47)
ADMA Per SD 1.12 (0.99-1.27)
FFA Quartile 4 vs 1 1.15 (0.78-1.70)
Phosphate Per SD 1.07 (0.94-1.23)
Marker Contrast [gamma]
CETP (b) Quartile 1 vs 4 0.642
HDL cholesterol Per SD 0.689
LDL cholesterol Per SD 0.733
sPLA2-lla mass Per SD (log) 0.809
Lp-PLA2 mass Tertile 3 vs 1 0.816
Uric acid Quartile 4 vs 1 0.879
CRP Per SD (log) 0.908
Cystatin C Quintile 5 vs 1 0.921
Triglycerides Per SD 0.928
SDMA Per SD 0.936
Cardiac troponin T Quartile 4 vs 1 0.943
A-FABP Per SD (log) 0.965
Creatinine clearance <60 vs 90+ mL/min 0.968
NT-proBNP Quartile 4 vs 1 0.976
F2RL3 [C] methylation Per SD 0.976
Calcium Per SD 0.980
PTH Per SD 1.001
MR-proANP Per SD (log) 1.004
GGT Quartile (d) 4 vs 1 1.006
ADMA Per SD (e)
FFA Quartile 4 vs 1 (e)
Phosphate Per SD (e)
(a) Sorted according to the trend of attenuation y; for details,
see text.
(b) CETP, cholesteryl ester transfer protein; Lp-PLA2,
lipoprotein-associated phospholipase A2; SDMA, symmetric
dimethylarginine; A-FABP, adipocyte fatty acid-binding protein.
(c) Human gene: F2RL3, F2R like thrombin/trypsin receptor 3.
(d) Sex-specific quartiles.
(e) Attenuation coefficient considered undefined; for details, see
text.
Table 2. Distribution of disease severity indicators at baseline,
in the full cohort and in patients still at risk after several
years of follow-up. (a)
Characteristic Category Baseline 3 years
Affected vessels 3-4 508(100) 484(95.3)
2 317(100) 307(96.8)
0-1 318(100) 314(98.7)
Left ventricular Moderate-severe 243(100) 228(93.8)
function impairment None-mild 858(100) 837(97.6)
NT-proBNP Quartile 4 287(100) 270(94.1)
Quartile 3 288(100) 277 (96.2)
Quartile 2 291 (100) 287 (98.6)
Quartile 1 295(100) 290(98.3)
Cardiac troponin T Quartile 4 285(100) 267 (93.7)
Quartile 3 289(100) 280(96.9)
Quartile 2 294(100) 288(98.0)
Quartile 1 290(100) 286(98.6)
Characteristic 6 years 10 years
Affected vessels 451 (88.8) 391 (77.0)
298 (94) 270 (85.2)
302 (95) 270 (84.9)
Left ventricular 204 (84.0) 170(70.0)
function impairment 807(94.1) 722(84.1)
NT-proBNP 243 (84.7) 196(68.3)
262(91.0) 235(81.6)
280 (96.2) 259 (89.0)
280 (94.9) 257 (87.1)
Cardiac troponin T 246 (86.3) 187 (65.6)
262 (90.7) 241 (83.4)
275 (93.5) 260 (88.4)
279 (96.2) 256 (88.3)
(a) Data shown are n (row-%).
Table 3. Distribution of biomarkers at baseline, in the full cohort,
and in patients still at risk after several years of follow-up. (a)
Characterist Category Baseline
CETP Quartile 1 284(100)
Quartile 2 287(100)
Quartile 3 277 (100)
Quartile 4 284(100)
HDL cholesterol, mg/dL 39.7(39.0-40.4)
Quartile 1 321 (100)
Quartile 2 312 (100)
Quartile 3 289 (100)
Quartile 4 280 (100)
LDL cholesterol, mg/dL 101 (99.5-103)
Quartile 4 296(100)
Quartile 3 289(100)
Quartile 2 304(100)
Quartile 1 297(100)
Lp-PLA2 mass Tertile 3 384(100)
Tertile 2 403(100)
Tertile 1 377(100)
Uric acid Quartile 4 285(100)
Quartile 3 265(100)
Quartile 2 304(100)
Quartile 1 292 (100)
CRP Log (mg/L) 1.24(1.15-1.32)
Quartile 4 284(100)
Quartile 3 292 (100)
Quartile 2 294(100)
Quartile 1 292 (100)
sPLA2-IIa mass Log (ng/mL) 0.97(0.91-1.03)
Quartile 4 289(100)
Quartile 3 290(100)
Quartile 2 285(100)
Quartile 1 295(100)
Cystatin C Quintile 5 229(100)
Quintile 4 217(100)
Quintile 3 233(100)
Quintile 2 230(100)
Quintile 1 253(100)
Characterist 3 years 6 years
CETP 274 (96.5) 254 (89.4)
274 (95.5) 264 (92.0)
268 (96.8) 260 (93.9)
280 (98.6) 262 (92.3)
HDL cholesterol, mg/dL 39.9 (39.2-40.6) 39.9 (39.1-40.6)
307 (95.6) 288 (89.7)
299 (95.8) 283 (90.7)
281 (97.2) 270 (93.4)
277 (98.9) 262 (93.6)
LDL cholesterol, mg/dL 101 (99.2-103) 101 (99.4-103)
286 (96.6) 272 (91.9)
278 (96.2) 266 (92.0)
295 (97.0) 279 (91.8)
291 (98.0) 275 (92.6)
Lp-PLA2 mass 365 (95.1) 347 (90.4)
393 (97.5) 372 (92.3)
369 (97.9) 349 (92.6)
Uric acid 268 (94.0) 249 (87.4)
258 (97.4) 249 (94.0)
297 (97.7) 285 (93.8)
288 (98.6) 273 (93.5)
CRP 1.21 (1.12-1.30) 1.20(1.11-1.28)
266 (93.7) 247 (87.0)
283 (96.9) 268 (91.8)
286 (97.3) 272 (92.5)
290 (99.3) 279 (95.5)
sPLA2-IIa mass 0.95 (0.89-1.01) 0.95 (0.89-1.02)
271 (93.8) 257 (88.9)
282 (97.2) 274 (94.5)
278 (97.5) 255 (89.5)
291 (98.6) 277 (93.9)
Cystatin C 213(93.0) 191 (83.4)
211 (97.2) 198 (91.2)
231 (99.1) 223 (95.7)
221 (96.1) 215 (93.5)
249 (98.4) 239 (94.5)
Characterist 10 years
CETP 223 (78.5)
234 (81.5)
238 (85.9)
229 (80.6)
HDL cholesterol, mg/dL 39.8 (39.1-40.6)
253 (78.8)
258 (82.7)
236 (81.7)
231 (82.5)
LDL cholesterol, mg/dL 101 (99.1-103)
242 (81.8)
232 (80.3)
249 (81.9)
247 (83.2)
Lp-PLA2 mass 316(82.3)
321 (79.7)
312 (82.8)
Uric acid 213(74.7)
229 (86.4)
251 (82.6)
244 (83.6)
CRP 1.16(1.06-1.25)
204(71.8)
239 (81.8)
248 (84.4)
257 (88.0)
sPLA2-IIa mass 0.93 (0.86-0.99)
217(75.1)
242 (83.4)
234 (82.1)
252 (85.4)
Cystatin C 152 (66.4)
179 (82.5)
203 (87.1)
200 (87.0)
214 (84.6)
(a) Data shown are n (row-%)or mean (95% CI)
only for markers analyzed as continuous variables.
Table 4. Association of baseline variables with mortality during
different periods of follow-up, adjusted for multiple covariables
(see text).
HR (a) (95% CI)
in follow-up period
Characteristic Comparison Overall
Patients at risk -- 1204
Deaths 276
Disease severity
indicators
Affected vessels 3-4 vs 0-1 1.28(0.91-1.81)
Left ventricular Moderate-severe 2.29 (1.77-2.97)
function impairment vs no-mild
NT-proBNP Quartile 4 vs 1 2.80(1.86-4.23)
Cardiac troponin T Quartile 4 vs 1 2.89(1.88-4.43)
Miscellaneous
biomarkers
CETP Quartile 1 vs 4 1.13(0.78-1.64)
HDL cholesterol Per SD 0.93 (0.81-1.06)
LDL cholesterol Per SD 1.07(0.94-1.22)
Lp-PLA2 mass Tertile 3 vs 1 1.36 (0.97-1.90)
Uric acid Quartile 4 vs 1 1.81 (1.26-2.60)
CRP Per SD (log) 1.35(1.18-1.54)
sPLA2-lla mass Per SD (log) 1.18(1.03-1.34)
Cystatin C Quintile 5 vs 1 1.98(1.35-2.88)
Characteristic Comparison Years 1-3
Patients at risk -- 1204
Deaths 35
Disease severity
indicators
Affected vessels 3-4 vs 0-1 5.16(1.18-22.6)
Left ventricular Moderate-severe 3.06(1.52-6.15)
function impairment vs no-mild
NT-proBNP Quartile 4 vs 1 4.61 (1.26-16.9)
Cardiac troponin T Quartile 4 vs 1 6.23(1.34-28.9)
Miscellaneous
biomarkers
CETP Quartile 1 vs 4 3.78(0.96-14.9)
HDL cholesterol Per SD 0.56(0.37-0.83)
LDL cholesterol Per SD 1.41 (0.99-2.01)
Lp-PLA2 mass Tertile 3 vs 1 2.87 (1.17-7.08)
Uric acid Quartile 4 vs 1 3.89(1.05-14.4)
CRP Per SD (log) 2.05(1.38-3.03)
sPLA2-lla mass Per SD (log) 1.52 (1.03-2.24)
Cystatin C Quintile 5 vs 1 4.95(1.05-23.4)
Characteristic Comparison Years 4-6
Patients at risk -- 1165
Deaths 53
Disease severity
indicators
Affected vessels 3-4 vs 0-1 0.97(0.46-2.02)
Left ventricular Moderate-severe 3.27(1.85-5.79)
function impairment vs no-mild
NT-proBNP Quartile 4 vs 1 2.68(1.04-6.87)
Cardiac troponin T Quartile 4 vs 1 3.41 (0.95-12.3)
Miscellaneous
biomarkers
CETP Quartile 1 vs 4 0.94 (0.44-2.04)
HDL cholesterol Per SD 0.97(0.73-1.29)
LDL cholesterol Per SD 0.88 (0.64-1.20)
Lp-PLA2 mass Tertile 3 vs 1 0.90 (0.43-1.89)
Uric acid Quartile 4 vs 1 1.04 (0.46-2.38)
CRP Per SD (log) 1.09 (0.80-1.48)
sPLA2-lla mass Per SD (log) 0.77(0.57-1.05)
Cystatin C Quintile 5 vs 1 1.88 (0.79-4.46)
Characteristic Comparison Years 7-1 0
Patients at risk -- 1104
Deaths 88
Disease severity
indicators
Affected vessels 3-4 vs 0-1 1.12 (0.64-1.98)
Left ventricular Moderate-severe 1.56 (0.97-2.51)
function impairment vs no-mild
NT-proBNP Quartile 4 vs 1 2.93(1.40-6.11)
Cardiac troponin T Quartile 4 vs 1 2.73(1.33-5.59)
Miscellaneous
biomarkers
CETP Quartile 1 vs 4 1.05 (0.56-1.97)
HDL cholesterol Per SD 1.10(0.88-1.37)
LDL cholesterol Per SD 1.08 (0.85-1.36)
Lp-PLA2 mass Tertile 3 vs 1 1.21 (0.64-2.27)
Uric acid Quartile 4 vs 1 1.50 (0.82-2.74)
CRP Per SD (log) 1.43(1.12-1.82)
sPLA2-lla mass Per SD (log) 1.30(1.03-1.64)
Cystatin C Quintile 5 vs 1 2.15(1.09-4.24)
Characteristic Comparison Years 11-13
Patients at risk -- 979
Deaths 100
Disease severity
indicators
Affected vessels 3-4 vs 0-1 1.18(0.65-2.13)
Left ventricular Moderate-severe 2.36(1.52-3.67)
function impairment vs no-mild
NT-proBNP Quartile 4 vs 1 2.28(1.17-4.43)
Cardiac troponin T Quartile 4 vs 1 2.26(1.16-4.41)
Miscellaneous
biomarkers
CETP Quartile 1 vs 4 1.03(0.53-2.00)
HDL cholesterol Per SD 0.88(0.69-1.12)
LDL cholesterol Per SD 1.09(0.88-1.35)
Lp-PLA2 mass Tertile 3 vs 1 1.37(0.79-2.38)
Uric acid Quartile 4 vs 1 2.38(1.25-4.51)
CRP Per SD (log) 1.26(1.01-1.57)
sPLA2-lla mass Per SD (log) 1.25(1.00-1.56)
Cystatin C Quintile 5 vs 1 1.59(0.89-2.86)
a HR, hazard ratio.
Table 5. Impact of taking into account updated biomarker values
measured during follow-up as time-varying predictors in
extended Cox regression models for general mortality, adjusted
for multiple covariables.
HR (a) (95% CI) in follow-up period
Characteristic Years 1-3 (BL) (b) Years 4-6 (BL)
NT-proBNP 4.61 (1.26-16.9) 2.68(1.04-6.87)
Cardiac troponin T 6.23(1.34-28.9) 3.41 (0.95-12.3)
HDL cholesterol 0.56(0.37-0.83) 0.97 (0.73-1.29)
LDL cholesterol 1.41 (0.99-2.01) 0.88(0.64-1.20)
CRP 2.05(1.38-3.03) 1.09(0.80-1.48)
Cystatin C 4.95(1.05-23.4) 1.88 (0.79-4.46)
Characteristic Years 4-6 (FU3) Overall (BL)
NT-proBNP 10.7 (2.74-41.6) 2.80 (1.86-4.23)
Cardiac troponin T 5.10 (1.27-20.5) 2.89 (1.88-4.43)
HDL cholesterol 0.79 (0.58-1.08) 0.93 (0.81-1.06)
LDL cholesterol 0.81 (0.62-1.06) 1.07 (0.94-1.22)
CRP 2.03 (1.52-2.72) 1.35 (1.18-1.54)
Cystatin C 7.14c (2.61-19.5) 1.98 (1.35-2.88)
Characteristic Overall (BL, FU3) Overall
(BL, FU1, FU3)
NT-proBNP 5.76(4.04-8.21) 8.08(5.68-11.5)
Cardiac troponin T 4.68(3.21-6.84) 4.86 (3.36-7.03)
HDL cholesterol 0.87 (0.77-1.00) 0.85 (0.74-0.96)
LDL cholesterol 0.87 (0.78-0.98) 0.90(0.81-1.01)
CRP 1.65(1.43-1.90) 1.80(1.57-2.06)
Cystatin C 4.03 (2.86-5.68) 4.66 (3.32-6.54)
(a) Hazard ratios (HRs) refer to the same comparisons as in Table 4,
using quantile cutoffs and SDs from the respective baseline
distributions.
(b) BL, baseline measurement; FU, measurement at follow-up
after 1 year; FU3, measurement at follow-up after 3 years.
(c) Model suffering from overspecification with some other HRs
converging to 0.
 Printer friendly
 Cite/link
 Email
 Feedback
| |
| Title Annotation: | Lipids, Lipoproteins, and Cardiovascular Risk Factors |
|---|---|
| Author: | Breitling, Lutz P.; Mons, Ute; Hahmann, Harry; Koenig, Wolfgang; Rothenbacher, Dietrich; Brenner, He |
| Publication: | Clinical Chemistry |
| Article Type: | Report |
| Date: | Mar 1, 2017 |
| Words: | 8121 |
| Previous Article: | Peripheral Inflammatory Biomarkers for Myocardial Infarction Risk: A Prospective Community-Based Study. |
| Next Article: | Cardiac Troponin T: Smaller Molecules in Patients with End-Stage Renal Disease than after Onset of Acute Myocardial Infarction. |
| Topics: | |