Study population

The study population came from the Uppsala Longitudinal Study of Adult Men (ULSAM), which was initiated in 1970, when all men born between 1920 and 1924 living in Uppsala, Sweden, were invited to a health survey (www.pubcare.uu.se/ULSAM). The present analyses were based on the baseline examination when participants were approximately 71 years of age. This population has thereafter been followed for a median of 9.8 years (range 0.1 -12.4 years). Of the 1221 participants, 940 had baseline plasma samples available for simultaneous measurements of biochemical markers. All participants gave written informed consent, and the ethics committee at the Faculty of Medicine of Uppsala University approved the study.

Baseline measurements

Information on clinical history and smoking status (current smoker vs. non-smoker) was obtained from a questionnaire. Participants’ smoking habits, body weight, body mass index (BMI), and waist circumference was obtained at the baseline visit. Obesity was defined as BMI ≥30 kg/m². Systolic and diastolic blood pressures were measured, and a 12-lead electrocardiogram was obtained with the participant in a supine position. Plasma glucose (fasting and 120 minutes after an oral glucose load) and fasting serum total, LDL and HDL cholesterol levels were measured by routine laboratory analyses. Type 2 diabetes mellitus was defined by fasting plasma glucose >7 mmol/L (corresponding to >126 mg/dL), or the use of oral hypoglycemic agents or insulin.

Biochemical methods

For biomarker measurements, venous blood samples were drawn at baseline and stored at –70°C for a median of 16.5 years (range 14.8-18.5) prior to analysis. CRP was assayed with the use of latex-enhanced reagents (Siemens), on a BN ProSpec analyzer (Siemens). The high sensitivity Troponin T (Lot number 153 401), NT-proBNP, GDF-15 and Cystatin C analyses were determined with sandwich immunoassays on Cobas Analytics immunoanalyzers (Roche Diagnostics). GDF-15 was measured with a pre-commercial assay (Roche Diagnostics) using a monoclonal mouse antibody for capture and a monoclonal mouse antibody fragment (F(ab´)2) for detection in a sandwich assay format. Detection was based on an electrochemiluminescence immunoassay (ECLIA), using a ruthenium(II) complex label. The pre-commercial assay correlates closely with a previously established IRMA method [10] (r=0.98, regression Passing/Bablok: slope 1.049, intercept -136 ng/L). The assay has an inter-assay coefficient of variation of 2.3% at 1100 ng/L and 1.8% at 17200 ng/mL, an intra-assay coefficient of variation of 0.8% at 1100 ng/L and 0.9% at 18600 ng/mL, and a lower detection limit of 10 ng/L.

Cardiovascular and cancer mortality and morbidity

Cardiovascular disease (CVD) was defined by the following criteria: prior myocardial infarction or angina pectoris, Q or QS waves or left bundle-branch block (Minnesota codes 1.1 to 1.3 and 7.1, respectively) on the baseline electrocardiogram; current treatment with nitroglycerin or cardiac glycosides, or a history of any CVD, as noted in the national registries. Complete information on mortality and morbidity from all patients were collected from the Swedish Cause of Death and Hospital Admission registries. The events were classified as CVD mortality (ICD-10 codes I00-I99), coronary heart disease (CHD) mortality or morbidity (ICD-10 codes I20-I25), stroke mortality or morbidity (ICD-10 codes I60-I69). Cancer disease was defined as any diagnosis of malignancy in the same registries (ICD-10 codes C00-D48). Concerning cancer, only the first non-fatal event was recorded; accordingly, cancer morbidity after the baseline visit only could be evaluated in participants without any previous cancer diagnosis at baseline. All clinical endpoints were classified as either previous or incident disease at the baseline visit or as occurring during follow-up.

Statistics

The study aimed at investigating the relations between GDF-15 and previous (cross-sectional) and subsequent (longitudinal) manifestations of cardiovascular and cancer disease, and mortality, both in the total population (n=940), and in subgroups without CVD (n=561) and without cancer disease (n=882) at baseline. The aims of the cross-sectional analyses were to investigate the relations between GDF-15 and established risk factors for the previous occurrence of CVD and cancer until the age 71. The aims of the longitudinal analyses were to investigate if GDF-15 provided incremental value beyond established risk factors and/or other biomarkers in predicting CVD and cancer during long-term follow-up.

Continuous variables were described with means and standard deviations. For continuous variables the Shapiro-Wilk’s test statistic W was calculated where the region W ≥0.95 with a respective original or logarithmic scale led to the use of a parametric method while otherwise a non-parametric method was used. All statistical tests and confidence intervals were two-sided (where applicable), and the region p<0.05 was used to declare a statistically significant result without adjustments for multiplicity. The univariate associations between the continuous risk factors (including other biomarkers) and GDF-15 were assessed with the parametric Pearson correlation, if both variables were normally distributed possibly after logarithmic transformation, and otherwise with the non-parametric Spearman rank correlation coefficients. The univariate associations between co-morbidities and GDF-15, and associations adjusting for cardiovascular risk factors, were assessed with logistic regression models. Results from the logistic regression models were presented as odds ratios, with 95% confidence intervals, of a one standard deviation (SD) increase of log GDF-15, and p values.

The linearity of the relation between continuous predictors and the longitudinal outcome events were examined visually in GAM plots and, in case of a non-normal distribution or non-linear relation, a logarithmic transformation was used. Proportional-hazards assumptions of Cox regression models were confirmed with Schoenfeld’s test. The longitudinal analyses were investigated with Cox proportional-hazards regression models and presented, for each continuous predictor, as hazard ratios, with 95% confidence intervals, of a one standard deviation increase of the predictor, and p values. Discriminative abilities of the models were estimated as C statistics for Cox regression models according to Pencina [31]. The increased discriminative ability of one regression model vs. another model was estimated based on the difference between two models in the individual estimated probability from logistic regression models using as measure the integrated discrimination improvement (IDI) and the continuous net reclassification index (NRI) according to Pencina [32,33]. The used Cox regression models included as predictors each of the following variables alone or in several different combinations: established risk factors (age, current smoking, BMI, total cholesterol, HDL cholesterol, lipid-lowering treatment, systolic blood pressure, antihypertensive treatment, type 2 diabetes, cancer before or at age 71), troponin T, GDF-15, NT-proBNP, CRP, and cystatin C. Kaplan-Meier curves (showing one minus event probability) were also presented stratified on tertiles of the biomarker.

Clinical characteristics and biomarkers at baseline

Clinical characteristics, concomitant diseases, the average levels of conventional risk factors, GDF-15, and other tested biomarkers in the total material of 940 men at the baseline investigation, are presented in Table 1. As expected there were slightly lower average levels of all risk factors and biomarkers in the 561 men without previous manifestation of CVD. The baseline characteristics in the 882 men without previous cancer were similar to the total population.

GDF-15 and its relations to biomarkers and co-morbidities at baseline

In the total population, the median GDF-15 level was 1494 ng/L (25^th and 75^th percentiles, 1216-1882 ng/L). The associations between baseline characteristics and tertiles of GDF-15 levels are shown in Table 2, and significances of the adjusted correlations in Table 3 and table 4. There were significant independent associations between GDF-15 levels and current smoking, diabetes mellitus, and biomarkers indicating renal dysfunction (cystatin C), cardiac dysfunction (NT-proBNP, troponin T) and also inflammatory activity (CRP). There were also significant independent relations between GDF-15 and previous cardiovascular or coronary heart disease and stroke morbidity, but not with previous cancer.

Relations to long-term total mortality

During follow-up there occurred 265 deaths in the total cohort, 131 in patients without CVD at baseline. Higher GDF-15 levels were significantly and log-linearly related to total mortality (Figure 1). The events were accrued at a fairly stable rate over the ten years follow-up (Figure 2). Per one SD increase in the level of log GDF-15, and after adjustment for conventional cardiovascular risk factors, an increase in total mortality by 48% (95%CI 33 to 67%, p<0.001) and 61% (95%CI 38 to 89%, p<0.001) were observed in the total cohort and the sample without CVD at baseline, respectively, which remained significant also after adjustment for other biomarkers (Table 5). When comparing the impact of the different biomarkers on total mortality, GDF-15 appeared to be a consistent independent prognostic marker of total mortality in the total population as well as in those without CVD or cancer disease at baseline (Figure 1). Also when evaluating the incremental prognostic value for mortality by addition of GDF-15 to conventional risk factors and other biomarkers there was a significant improvement of the integrated discrimination index (IDI) and the net reclassification index (NRI) (Table 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. GDF-15 and other biomarkers in relation to cardiovascular and cancer outcomes.

Adjusted hazard ratios with 95% CI for 1 SD increase of log transformed GDF-15 values and of log transformed values of other biomarkers in relation to 10 years outcome of total mortality, cardiovascular disease (CVD) mortality, CVD mortality or coronary heart disease (CHD) or stroke morbidity/mortality, cancer mortality and cancer morbidity/mortality, in the whole sample (n = 940), subjects without cardiovascular disease (non CVD sample, n = 561) and subjects without cancer disease (non cancer sample, n = 882) at baseline (outcome cancer morbidity/mortality only in the non-cancer sample). Hazard ratios are adjusted for age, current smoking, body mass index, systolic blood pressure, antihypertensive treatment, total cholesterol, HDL cholesterol, lipid-lowering treatment, type 2 diabetes, and cancer at baseline (except in population P3), and log transformed values of the biomarkers GDF‑15, NT-proBNP, troponin T, cystatin C, and CRP.

https://doi.org/10.1371/journal.pone.0078797.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Relations between GDF-15 levels and cardiovascular and cancer outcomes.

GDF-15 (ng/l) in relation to logit of risk for total mortality, cardiovascular disease (CVD) mortality, coronary heart disease (CHD) morbidity/mortality, stroke morbidity/mortality, CVD mortality or CHD or stroke morbidity/mortality combined, and cancer mortality.

https://doi.org/10.1371/journal.pone.0078797.g002

Relations to long-term cardiovascular morbidity and mortality

There were 304 cases with cardiovascular or stroke morbidity or mortality events, including 115 cardiovascular deaths during follow-up out of which 147 respectively 41 cases occurred in the non-cardiovascular disease subgroup. In the total population as well as in the non-CVD population an increasing GDF-15 level was significantly and log-linearly related to total and CVD mortality as well as to CHD and stroke morbidity and mortality and to cancer mortality (Figure 2). The events were accrued at a fairly stable rate over the ten years follow-up (Figure 3). After adjustment for conventional cardiovascular risk factors, one SD increase in the level of log GDF-15 were, in the respective total and non-CVD populations, associated with 48% (95%CI 26 to 73%, p<0.001) and 67% (95%CI 28 to 217%, p<0.001) incremental risk of cardiovascular mortality and 38% (95%CI 24 to 53%, p<0.001) and 47% (95%CI 25 to 73%, p<0.001) increases in CVD mortality and CHD or stroke morbidity and mortality after adjustment for conventional cardiovascular risk factors (Table 5). After adjustment also for other biomarkers the incremental rise in outcome events by one SD increase of log GDF-15 remained significant for all events except stroke. When comparing the impact of the different biomarkers on outcomes in the total population and in cohorts without CVD at entry, GDF-15 was a consistent independent prognostic marker concerning all events (Figure 1). Also when evaluating the incremental prognostic value for cardiovascular morbidity and mortality by addition of GDF-15 to conventional risk factors and other biomarkers, there was a significant improvement of the integrated discrimination index (IDI) and the net reclassification index (NRI) (Table 6).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Kaplan-Meier estimates of risk of cardiovascular and cancer outcomes in relation to GDF-15.

Kaplan-Meier estimates of event-free probability functions by tertiles of GDF-15 in the total material of 940 subjects for total mortality; cardiovascular disease (CVD) mortality; coronary heart disease (CHD) morbidity or mortality; stroke morbidity or mortality; CVD mortality or CHD or stroke morbidity/mortality; and cancer mortality.

https://doi.org/10.1371/journal.pone.0078797.g003

Relations to long-term cancer morbidity and mortality

During follow-up, out of the 940 subjects in the total cohort, 105 deaths were caused by cancer. In the group without cancer at entry (n=882), 182 cases on new cancer morbidity were found, including 85 deaths caused by cancer. During the follow up in the total population as well as in the non-cancer-population an increasing GDF-15 level was significantly related to subsequent cancer mortality (Figure 2). The cancer events accrued at a fairly stable rate over the ten years follow-up (Figure 3). By one SD increase of the level of log GDF-15 there was in the total population a significant 36% (95%CI 15 to 62%, p<0.001) increase in cancer mortality and in the population without cancer disease at baseline an increase of 37% (95%CI 13 to 66%, p<0.001) in cancer mortality and 26% (95%CI 10 to 45%, p<0.001) in cancer morbidity and mortality (Table 7). After adjustments for conventional risk factors, the incremental rise in outcome events by one SD of log GDF-15 remained significant 46% (95%CI 21 to 77%, p<0.001) and 38% (95%CI 12 to 70%, p<0.001) for cancer mortality and 30% (95%CI 12 to 51%, p<0.001) for cancer morbidity and mortality in the respective groups. Only GDF-15 independently predicted cancer morbidity and mortality in the context of all biomarkers (Figure 1). When comparing the impact of the different biomarkers on outcomes, CRP and GDF-15 were the only biomarkers related to cancer morbidity and mortality. When also including the other available baseline characteristics and biomarkers only GDF-15 was significantly related to cancer morbidity and mortality (Figure 1). Sensitivity analyses using lag times of one or two years respectively before counting cancer events gave similar results with maintained statistical significances of results observed. Finally when evaluating the incremental value by addition of GDF-15 to conventional risk factors there was a significant improvement of integrated discrimination index (IDI) and the net reclassification index (NRI) although the significances were somewhat weakened when added also to information from all other biomarkers (Table 8).