Competing risks and lifetime coronary heart disease incidence during 50 years of follow-up

    Competing risks and lifetime coronary heart disease incidence during 50 years of follow-up Paolo Emilio Puddu, Paolo Piras, Alessandro Menotti PII: DOI: Reference:

S0167-5273(16)30949-4 doi: 10.1016/j.ijcard.2016.05.043 IJCA 22599

To appear in:

International Journal of Cardiology

Received date: Accepted date:

1 April 2016 12 May 2016

Please cite this article as: Puddu Paolo Emilio, Piras Paolo, Menotti Alessandro, Competing risks and lifetime coronary heart disease incidence during 50 years of follow-up, International Journal of Cardiology (2016), doi: 10.1016/j.ijcard.2016.05.043

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

1

SC R

IP

Paolo Emilio Puddu (1), Paolo Piras (1), Alessandro Menotti (2)

T

Competing risks and lifetime coronary heart disease incidence during 50 years of follow-up

MA

NU

(1) Department of Cardiovascular, Respiratory, Nephrological, Anesthesiological and Geriatric Sciences, Sapienza University of Rome, Rome, Italy (2) Association for Cardiac Research, Rome, Italy

TE

D

Running Head: Competing risks and lifetime incidence of coronary heart disease

Title: 13

CE P

Word Counts:

Abstract: 237

AC

Manuscript: 2786 (exclusive of References) References: n=22 (572 words) Tables: 2

Figures: 2 Appendix: 319

Author for correspondence and reprints: P.E. Puddu, MD, PhD, FESC, FACC, Laboratory of Biotechnologies Applied to Cardiovascular Medicine, Department of Cardiovascular, Respiratory, Nephrological, Anesthesiological and Geriatric Sciences, Sapienza University of Rome, Viale del Policlinico, 155, Roma 00161, Italy. Tel. +39.06.49972659; Fax. +39.06.4453891; e-mail: [email protected]

ACCEPTED MANUSCRIPT

2

Abstract Objectives: To study coronary heart disease (CHD) incidence versus other cause of death

IP

the differential role of risk factors for different end-points.

T

using the cumulative incidence function and the competing risks procedures to disentangle

SC R

Material and methods: We compared standard Cox and Fine-Gray models among 1677 middle aged men of an Italian population study of cardiovascular diseases that reached 50

NU

years of follow-up with the quasi extinction of the population. The incidence of either fatal or non-fatal cases in 50 years was used as primary event, while deaths from any other cause,

considered 10 selected risk factors.

MA

mutually exclusive from the primary events, were considered as secondary events. We

D

Results: The main result was that cholesterol was significantly and positively related to

TE

incidence of CHD contrasted with deaths from any other cause. On the other hand, when the primary events were deaths from any other cause and the competing events were CHD,

CE P

cholesterol was inversely and age positively related. This outcome did not exclude the predictive role of other risk factors, such as age, cigarettes, arm circumference (protective),

AC

systolic blood pressure, vital capacity (protective), cholesterol, corneal arcus and diabetes, documented by the Cox model, that had common roles for both end-points. Conclusions: Fine-Gray model, initially proposed to handle adequately cumulative incidence function may thus prevent overestimation of risks related to the Kaplan-Meyer based methods such as Cox model and identify the specific risk factors for defined end-points.

Key words: Predictive models; Cox; Fine-Gray; competing risks; CHD typical; cholesterol; Risk factors; Epidemiology; 50-year follow-up

ACCEPTED MANUSCRIPT

3

Introduction Some population studies of long duration, started in the last century, have reached the

T

stage of extinction or quasi-extinction of the populations themselves. This allows, under some

IP

circumstances, to estimate morbid events during lifetime [1-7]. The evaluation of rates, risk

SC R

and predictors (determinants, risk factors) in this situation may produce difficulties in handling and interpretation of data and findings. In fact, it is well known that the simple

NU

Kaplan-Meier survival curves are distorting reality since they tend to overestimate the risk and reduce survival mainly when the follow-up in very long. Intriguing interpretations may

MA

arise from these findings when dealing with standard multivariable predictive models as they do not take into account the role of the so-called competing risks, that is the effect of morbid

D

and/or fatal conditions that are alternative (and in competition) with the basic studied

TE

condition [8-11].

Although special procedures were proposed to estimate survival correctly and to make

CE P

prediction by appropriate models [9,10], these were rarely applied since investigations that reached the extinction or the quasi-extinction of the study populations are rare. We thus

AC

attempted to use these novel approaches in an Italian population study of cardiovascular diseases that reached 50 years of follow-up with the quasi extinction of the population. The incidence of coronary heart disease (CHD) was used as primary event, while deaths from any other cause, mutually exclusive from the primary events, were considered as secondary events.

ACCEPTED MANUSCRIPT

4

Material and Methods The two Italian Rural Areas (IRA) of the Seven Countries Study (SCS) of

T

Cardiovascular Diseases were considered for this analysis. They were enrolled and first

IP

examined in 1960 and made by a total of 1712 men aged 40-59 representing 98.5% of

SC R

defined samples [7].

Risk factors. A selected group of risk factors were considered as follows: a) age

NU

(years) approximated to the nearest birthday; b) cigarette smoking (n/day) derived from a standard questionnaire; c) body mass index (units) computed from height and weight,

MA

measured following the technique described in the World Health Organization (WHO) Cardiovascular Survey Methods Manual [12] (WHO Manual); d) arm circumference (mm)

D

measured at right arm following the technique described in the WHO Manual [12] and

TE

mathematically cleaned from the bicipital skinfold thickness; e) systolic blood pressure (mmHg) measured at right arm, in supine position following the technique described in the

CE P

WHO Manual [12] using the average of two measurements; f) heart rate (beats/min) derived from a resting ECG tracing; g) vital capacity (l/m2) following the technique described in the

AC

WHO Manual [12] and using the best of three attempts; h) serum cholesterol (mmol/L) measured on casual blood sample following the technique of Anderson and Keys [13]; i) corneal arcus (present-absent) as judged by a physician; l) diabetes (present-absent) derived from history, possible specific treatment and urine glucose measurement. End-points. Incidence of CHD in 50 years of follow-up was considered the primary end-point and was measured exploiting all possible information collected at baseline, at periodical re-examinations, at special search at hospitals, clinics and general practitioners, interviews with relatives and data from causes of death as described in detail elsewhere [14,15]. Diagnoses were based on history, physical examination, ECG tracings, occasionally reported diagnoses, and causes of death. CHD were cases manifested as sudden death (when

ACCEPTED MANUSCRIPT

5

other causes could be reasonably excluded), fatal and non fatal myocardial infarction, other fatal and non fatal acute ischemic syndromes. Heart disease manifested only as heart failure,

T

severe chronic arrhythmia, heart blocks, documented diagnoses of hypertensive heart disease

IP

or “chronic CHD” were not classified as CHD for reasons given elsewhere [14,15]. In

SC R

previous analyses these cases were classified as Atypical CHD or Heart Disease of Uncertain Etiology since they had not relationship with serum cholesterol. In the present analyses these cases are incorporated into the group of death from other causes or of survivors.

NU

Incidence cases were associated with a date corresponding to the first event occurred in

MA

50 years. There were 35 cases of definite or possible CHD cases at entry examination and these prevalent cases were excluded from analysis.

D

Cases of death from any other cause in 50 years among those who remained free from

TE

CHD were considered as secondary (competing) end-point (OTHER DEATHS). Collection of mortality data along 50 years was complete and, beyond the availability of death

CE P

certificates, it was largely based on a procedure that anticipated in principle and content of the so called WHO Verbal Autopsy [16]. Causes of death, accompanied by the respective

AC

date, were coded by the WHO ICD-8 [17] and based on defined criteria. Baseline data were collected before the era of the Helsinki declaration. Subsequently verbal consent was obtained in view of collecting and using follow-up information. Statistical analysis. The denominator was of 1677 units. Each individual could suffer none, one or more CHD events but only the first event (either fatal or non fatal) with its date of occurrence was used for analysis. Kaplan–Meier survival curves related to CHD or death from any other cause were computed. Mean values of risk factors at entry examination were computed. Cox proportional hazards models were solved with 50-year CHD incidence as end-points and the risk factors measured at entry as predictors. In this case those surviving or

ACCEPTED MANUSCRIPT

6

dying from other cause in 50 years were considered a censored. Another Cox model was solved using as end-point the combination of CHD event and OTHER DEATHS.

T

Similar models were solved for OTHER DEATHS playing the role of competing risks

IP

versus CHD incidence in 50 years, and vice-versa by using Fine-Gray model elaborated for

SC R

proportional hazards with the subdistribution of a competing risk [9] using the R package as described by Gray [10]. Therefore CHD events and OTHER DEATHS were alternatively the principal event and the competing event. Another end-point was made by the combination of

MA

NU

CHD events plus OTHER DEATHS and was called Combined Events (COMB).

D

Results

TE

In 50 years, among the 1677 men CHD-free at entry examination, there were 451 CHD

CE P

events, 1190 OTHER DEATHS and 36 survivors. The average values of all 10 covariates measured at baseline among the 1677 men are illustrated in Table 1 according to the distribution in Survivors, CHD events and OTHER

AC

DEATHS. There were some differences among the 3 groups. Significantly different levels were found for 6 risk factors comparing Survivors with CHD incidence; for 5 risk factors comparing Survivors with OTHER DEATHS; and for 4 risk factors comparing CHD incidence with OTHER DEATHS. Figure 1 illustrates the Kaplan-Meier survival curves of CHD incident cases and OTHER DEATHS during 50 years of follow-up. The survival for both end-points was very small confirming that the Kaplan Meier approach tends to overestimate the risk. Table 2 shows the solutions of Cox and Fine-Gray models. In the first Cox model, risk factors significantly predicting events were age, serum cholesterol, cigarette smoking, corneal arcus, and systolic blood pressure, while vital capacity was not far from significant levels

ACCEPTED MANUSCRIPT

7

confirming notions already documented elsewhere. In the Cox model with combined events (COMB) as end-point 8 of 10 covariates were significantly related to the end-point: age,

T

cigarettes, arm circumference (protective), systolic blood pressure, vital capacity (protective),

IP

cholesterol, corneal arcus and diabetes.

SC R

The Fine-Gray models were characterized by a single end-point, i.e. COMB including 2 types of events (CHD and OTHER DEATHS), treated alternatively as principal or competing event. Therefore, the solutions were different depending on whether CHD was the primary or

NU

competing event. Serum cholesterol was the only risk factor significantly and positively

MA

related to CHD as primary event, when OTHER DEATHS played the role of competing risk while when CHD competed versus OTHER DEATHS the significant factors were age and

D

cholesterol, the latter interestingly with a negative (inverse relation) coefficient.

TE

The findings concerning the differential role of risk factors can approximately be obtained using another simpler- almost trivial- procedure. In fact, taking into account two

CE P

Cox models, one with CHD as end-point versus all other cases, and the other one with OTHER DEATHS versus all other cases, a t test applied to compare all pairs of coefficients

AC

of the various risk factors, show that the only significant differences are those related to serum cholesterol and age. Figure 2 shows, in the left-hand panel, the Fine-Gray cumulative incidence functions (CIF) of OTHER DEATHS and CHD incidence. CIF at time t, for each one of these types of events, is the probability of an event of that type occurring at any time point between baseline and time t. Since the data set contains a few censored observations (n=36), i.e. not all subjects were observed to experience an event, this estimate was properly modified to correctly account for censoring. As time increases, the CIF increases from zero to the total proportion of events of that type so that the sum of CIFs gives the total probability of events (n=1641 of 1677 or 0.979) in the data set, different to what is obtained by Kaplan-Meier whereby the

ACCEPTED MANUSCRIPT

8

sum of the probabilities is much larger than 1. It is evident that CIF is 2.64 times greater for OTHER DEATHS (0.710) than for CHD incident cases (0.269). Trying instead to extrapolate

T

an estimate of risk, from Figure 1 (dealing with the Kaplan Meier approach) it appears that,

IP

for both end-points, risk was much higher than that given in the left panel of Figure 2. The

SC R

Fine-Gray models whereby the competing event was, respectively, the OTHER DEATHS (Fig. 2 central panel: CHD as primary event) and CHD (Fig. 2 right panel: CHD as secondary event) are also illustrated. The main result concerns the panel where CHD is the primary

NU

event (Fig. 2 central panel). In this case we observe that the CHD curve is above the one of

MA

OTHER DEATHS and this is positively related to cholesterol as indicated in Table 2. The right panel derives from switching the two competing events. This is useful if one wants to compare the complementary roles of the two competing events [18]. In our case the

TE

D

secondary (competing) event is not specific, being “from any other deaths”. This means that the “competing” relationship is not symmetrical and that the coefficients of the analysis

CE P

where CHD is the secondary event are not directly interpretable. For example, in this case, a low cholesterol level increases the risk of OTHER DEATHS. In fact, CHD prediction line is

AC

below that of OTHER DEATHS. It is obvious that this statistical result should be coupled with the primary analysis. What is important is that in both analyses cholesterol is significant and suggests that, when OTHER DEATHS are treated in competition with CHD cholesterol itself is the primary risk factor for CHD.

ACCEPTED MANUSCRIPT

9

Discussion The results of this analysis were obtained by applying the Fine-Gray model that was

T

initially proposed to handle adequately CIF and thus might prevent overestimation of risks

IP

related to the Kaplan-Meyer based methods such as the Cox model [8-11]. The main finding

SC R

from the present prospective epidemiological investigation among 1677 middle-aged men enrolled in the 1960’ and followed-up for half a century, thus reaching the practical

NU

extinction of the initial cohort, was that cholesterol is significantly and positively related to the incidence of CHD when the competing condition was OTHER DEATHS. On the other

MA

hand, when the primary events were OTHER DEATHS and the competing events were CHD incidence, cholesterol was inversely and age positively related. Indeed, by the Cox model the

D

large majority of measured covariates were predictors of events, jointly considered, and there

TE

was not an opposite sign (and role) of the coefficient for cholesterol when CHD was competing versus OTHER DEATHS as primary events. Apparently this means that the

CE P

discriminant predictor between CHD events and OTHER DEATHS is serum cholesterol without excluding, on the other hand, the common predictive role of the other risk factors

AC

considered in this analysis. However, this role cannot be seen by using the Fine-Gray models. The longer the follow-up the higher the likelihood that other than the events considered of primary interest come to the scene to compete with the formers [11]. However, experience whereby follow-up lasts over 25-45 years are rare [1-7,14,15] and applications of competing risk modelling are definitely an occasional encounter due, at least in part, to the few computing tools available in current statistical packages [10,11]. The literature on the subject, as dealing the coronary or cardiovascular diseases, is rare. A paper of 2005 [19] tried to disentangle the role of some CHD risk factors between the occurrence of myocardial infarction, stroke and venous thrombo-embolism. Hypertension, high serum cholesterol, diabetes and smoking habits were associated with myocardial

ACCEPTED MANUSCRIPT

10

infarction and stroke but not with the third condition that was associated with body mass index and height. Two more recent contributions [20,21] were able to improve predictions of

T

events (CHD and stroke) by applying the competing risks procedure in risk functions

IP

produced for predictive purposes. However, the follow-up duration of these latter

SC R

investigations was relatively short, on average less than or equal to 20 years, whereas the age ranges of the populations studied at enrolment were larger (interval 40-90 years) [19-21]. Moreover, in two studies, the type of risks in competition was closer, all in all being of

NU

atherosclerotic nature (CHD versus stroke or CVD) [20,21]. It is possible that these

MA

differences versus the present investigation might explain why by Fine-Gray models the common predictive roles of some risk factors was still seen there in presence of competing

D

risks [20,21]. Our data and results, on the contrary, contrast the specific pathology under

TE

investigation (CHD) with a series of competing events that, all together, represent occurrences that compete largely with the aetiology of CHD, namely atherosclerosis.

CE P

The scarce literature on the issue contrasts with the importance of the scientific question that might accordingly be replied. As the probability of events of a given type is

AC

heavily influenced by concomitant events that might happen among the exposed individuals, how one may handle subjects who will not be censored but will have a different (from the interest) event and whom are expected to undergo a steep increase in numbers as follow-up becomes longer ? Our results clearly show that application of the Fine-Gray model based on CIF enables to dissect the respective role that baseline covariates may have to segregate the probabilities of two types of events in contrast from each other. They point to cholesterol as the factor positively related to the incidence of fatal and non-fatal CHD and negatively related to mortality from any other cause during a half a century follow-up among a middleage cohort of men. Moreover, age is more positively associated with mortality from any other

ACCEPTED MANUSCRIPT

11

cause than CHD. This suggests the critical role of serum cholesterol in discriminating the types of events.

T

It remains for further study in this material to investigate whether subdividing OTHER

IP

DEATHS into multiple causes of deaths may enable to sort out the role of other specific

SC R

risks. Insights from this material however point to the extensive role of cardiovascular risk factors also to increase the risk of other than CHD events [22]. It might be also of interest to assess whether just comparing mortality data (i.e. CHD deaths versus specific death types)

NU

might make a difference in addressing specific risk factors and/or whether this might

MA

disentangle better risk factors among those investigated. A limitation in this material will be however that a few risk factors were measured and the highest focus, by design [12-15], was

D

on cardiovascular ones.

TE

The evidence presented here calls for a multicenter effort to merge data of very longterm observations whereby risks might be challenged in competition using appropriate

CE P

modelling such as the Fine-Gray CIF based model [8,9] and its software applications [10]. Wolbers et al. have recently addressed objectives and approaches of competing risk analyses

AC

[11]. However, the cases presented where of very short duration and of quite difficult interpretation and applicability while the ideal condition, whereby one may wish to test competition of risks, is between hard mutually exclusive events and the optimum should be to contrast different types of deaths over a long follow-up. How large might be the difference between types of risks in competition is a further aspect that needs research and in depth testing as the problem of genders and of age ranges among the enrolled participants [19-21]. The reason why this demands a multicenter approach is that the follow-up may extend practically during all life span as in the present case and one needs a very large denominator which is likely to be accrued only by fusion of multiple data sets.

ACCEPTED MANUSCRIPT

12

It is important to conclude that when follow-up is of very long duration (say more than 25-40 years) the need exists to adopt predictive models capable of handling competing risks.

T

These considerations may well stimulate revisions of existing results whereby competing

AC

CE P

TE

D

MA

NU

SC R

IP

risks of CHD were not assessed.

ACCEPTED MANUSCRIPT

13

References 1.

Fraser GB, Lindsted KD, Beeson WI. Effect of risk factor values on lifetime risk and

T

age at first coronary event: the Adventist Health Study. Am J Epidemiol. 1995; 142:

Doll R, Peto R, Boreham J, Sutherland I. Mortality in relation to smoking: 50 years’

SC R

2.

IP

746-758.

observations on male British doctors. Br Med J 2004; 328: 1519-1531. Lloyd-Jones DM, Leip EP, Larson MG, et al. Prediction of lifetime risk for

NU

3.

cardiovascular disease by risk factor burden at 50 years of age. Circulation 2006; 113:

4.

MA

791-798.

Lloyd-Jones DM, Dyer A, Wang R, Daviglus ML, Greenland P. Risk factor burden in

D

middle age and lifetime risk for cardiovascular and non-cardiovascular death (Chicago

5.

TE

Heart Association Detection Project in Industry) Am J Cardiol. 2007; 99: 535-540. Ferrie JE, Singh-Manoux A, Kivimäki M, et al. Cardiorespiratory risk factors as

6.

CE P

predictors of 40-year mortality in women and men. Heart 2009; 95: 1250-1257. Tielemans SMAJ, Geleijnse JM, Menotti A, et al. Ten-year blood pressure trajectories,

AC

cardiovascular mortality, and life years lost in 2 extinction cohorts: the Minnesota business and professional men study and the Zutphen Study. J Am Heart Assoc 2015; 4: e001378. 7.

Menotti A, Puddu PE, Lanti M, Maiani G, Fidanza F. Cardiovascular risk factors predict survival in middle-aged men during 50 years. Eur J Intern Med 2013;24: 67-74.

8.

Grey RJ. A class of K-sample tests for comparing the cumulative incidence of a competing risk. Ann Stat 1988; 16: 1141-1154.

9.

Fine JP, Gray RJ. A proportional hazards model for the subdistribution of a competing risk. J Am Stat Assoc 1999; 94: 496-509.

10.

Gray B. 2014. Cmprsk R package v. 2.2-7.

ACCEPTED MANUSCRIPT 11.

14

Wolbers M, Koller MT, Stel VS, Schaer B, Jager KJ, Leffondré K, Heinze G. Competing risks analyses: objectives and approaches. Eur Heart J 2014; 35: 2936-2941. Rose G, Blackburn H. Cardiovascular Survey Methods. Geneva, 1968, World Health

T

12.

Anderson JT, Keys A. Cholesterol in serum and lipoprotein fractions: its measurement and stability. Clin Chem 1956; 2: 145-159.

14.

SC R

13.

IP

Organization.

Puddu PE, Menotti A. Natural history of coronary heart disease and heart disease of

NU

uncertain etiology: findings from a 50-year population study. Int J Cardiol 2015; 197:

15.

MA

260-264.

Menotti A, Puddu PE. Lifetime prediction of coronary heart disease and heart disease

D

of uncertain etiology in a 50-year follow-up population study. Int J Cardiol 2015; 196:

16.

TE

55-60.

WEB site. Verbal autopsy standards. The 2012 WHO verbal autopsy instruments.

17.

CE P

Release Candidate 1. WHO Geneva. Visited February 2015 World Health Organization. 8th International classification of diseases. 8th revision.

18.

Pintilie M. An introduction to competing risk analysis. Rev Esp Cardiol 2011; 64: 599605.

19.

AC

Geneva. World Health Organization: 1965.

Glynn RJ, Rosner B. Comparison of risk factors for the competing risks of coronary heart disease, stroke and thrombo-embolism. Am J Epidemiol 2005; 162: 975-982.

20.

Wolbers M, Koller MT, Witteman JCM, Steyerberg EW. Prognostic models with competing risks. Methods and application to coronary heart disease risk prediction. Epidemiology 2009; 20: 555-561.

21.

Van Kempen BJ, Ferket BS, Kavoui M et al. Performance of Framingham cardiovascular disease (CVD) predictions in the Rotterdam Study taking into account

ACCEPTED MANUSCRIPT

15

competing risks and disentangling CVD into coronary heart disease (CHD) and stroke. Int J Cardiol 2014; 171: 413-418.

T

Menotti A, Puddu PE, Maiani G, Catasta G. Cardiovascular and other causes of death

IP

as a function of lifestyle habits in a quasi extinct middle-aged male population. Int J

CE P

TE

D

MA

NU

SC R

Cardiol 2016; 210: 173-178.

AC

22.

ACCEPTED MANUSCRIPT

16

Table 1. Mean covariates values for the categories described in the text.

IP

T

CHD incidence mean SD 48.77 4.98 9.08 9.49 25.66 .63 270.83 22.73 143.63 20.00 71.22 12.73 1.65 0.25 5.37 1.07 0.149 0.356 0.049 0.216

NU

Age (years) Cigarettes (n/day) Body Mass Index (units) Arm circumference (mm) Systolic blood pressure (mmHg) Heart rate (beats/min) Vital capacity (l/m2) Cholesterol (mmol/l) Corneal arcus (%) Diabetes (%)

Survivors mean SD 42.92 2.45 4.53 7.93 24.76 2.61 274.2 19.24 130.22 11.52 66.47 9.05 1.81 0.18 4.87 0.94 0.028 0.167 0 -

SC R

Panel A

p 0.0001 0.0053 0.1457 0.3875 0.0001 0.0287 0.0002 0.0068 0.0446 0.4070

p <0.0001 0.0083 0.6426 0.1048 0.0002 0.0259 <0.0001 0.0987 0.0756 0.3014

D

MA

Survivors versus OTHER DEATHS

AC

CE P

Age (years) Cigarettes (n/day) Body Mass Index (units) Arm circumference (mm) Systolic blood pressure (mmHg) Heart rate (beats/min) Vital capacity (l/m2) Cholesterol (mmol/l) Corneal arcus (%) Diabetes (%)

Survivors versus CHD

TE

Panel B

SD: standard deviation

OTHER DEATHS mean SD 49.33 5.06 8.79 9.56 25.05 3.72 267.68 23.86 143.19 20.48 71.28 12.84 1.64 0.24 5.16 1.04 0.136 0.343 0.047 0.212

CHD versus OTHER DEATHS p 0.0402 0.5826 0.0029 0.0158 0.6958 0.9325 0.4585 0.0003 0.3347 0.8785

ACCEPTED MANUSCRIPT

17

Table 2. Cox proportional hazards models dealing with CHD and combined events (COMB),

TE

D

MA

Cox Model - CHD+OTHER DEATHS (COMB) 0.0854 5 1.53 1.45 1.62 0.0001 0.0163 10 1.18 1.12 1.24 0.0001 0.0034 3.5 1.01 0.95 1.08 0.7150 -0.0034 25 0.92 0.86 0.98 0.0162 0.0104 20 1.23 1.16 1.30 0.0001 0.0022 15 10.3 0.97 1.10 0.3181 -0.4873 0.25 0.89 0.84 0.94 0.0001 0.0865 1 1.09 1.04 1.14 0.0005 0.3085 1 1.36 1.18 1.57 0.0001 0.3115 1 1.37 1.08 1.72 0.0080

CE P

Age Cigarettes Body mass index Arm circumference Systolic blood pressure Heart rate Vital capacity Serum cholesterol Corneal arcus Diabetes

NU

SC R

IP

T

and Fine-Gray model dealing with COMB alternatively as principal and competing. coefficient delta Hazard ratio 95% CI P Cox Model - CHD Age 0.0404 5 1.22 1.10 1.36 0.0001 Cigarettes 0.0131 10 1.14 1.04 1.25 0.0073 Body mass index 0.0246 3.5 1.09 0.97 1.22 0.1446 Arm circumference -0.0014 25 0.97 0.85 1.10 0.5982 Systolic blood pressure 0.0087 20 1.19 1.07 1.32 0.0011 Heart rate -0.0012 15 0.98 0.87 1.11 0.7758 Vital capacity -0.3721 0.25 0.91 0.82 1.01 0.0911 Serum cholesterol 0.1743 1 1.19 1.09 1.30 0.0001 Corneal arcus 0.3467 1 1.41 1.08 1.85 0.0116 Diabetes 0.2252 1 1.25 0.81 1.93 0.3090

AC

Fine-Gray Model - CHD=principal, OTHER DEATHS= competing Age -0.0118 5 0.94 0.85 1.04 0.2400 Cigarettes 0.0043 10 1.04 0.95 1.15 0.3700 Body mass index 0.0232 3.5 1.08 0.96 1.22 0.1800 Arm circumference 0.0011 25 1.03 0.90 1.17 0.6700 Systolic blood pressure 0.0015 20 1.03 0.93 1.15 0.5800 Heart rate -0.0025 15 0.96 0.85 1.09 0.5600 Vital capacity -0.0701 0.25 0.98 0.88 1.09 0.7500 Serum cholesterol 0.1415 1 1.15 1.05 1.26 0.0021 Corneal arcus 0.1592 1 1.17 0.89 1.55 0.2600 Diabetes 0.0348 1 1.04 0.66 1.62 0.8800 Fine-Gray Model - OTHER DEATHS=principal, CHD = competing Age 0.0463 5 1.26 1.18 1.35 0.0000 Cigarettes 000055 10 1.06 0.99 1.13 0.1000 Body mass index -0.0134 3.5 0.95 0.88 1.03 -1.1480 Arm circumference -0.0030 25 0.93 0.85 1.01 -1.7840 Systolic blood pressure 0.0032 20 1.07 0.99 1.15 1.6860 Heart rate 0.0040 15 1.06 0.98 1.15 01400 Vital capacity -0.1259 0.25 0.97 0.90 1.04 0.3800 Serum cholesterol -0.0637 1 0.94 0.88 1.00 0.0400 Corneal arcus -0.0180 1 0.98 0.80 1.20 -0.1740 Diabetes 0.1236 1 1.13 0.84 1.52 0.8120

ACCEPTED MANUSCRIPT

18

T

LEGENDS

IP

Figure 1. Survival (Kaplan Meyer) curves for subjects referred to the two types of event:

SC R

CHD versus OTHER DEATHS.

NU

Figure 2. Cumulative incidence curves (left) and Fine-Gray models (center and right) where the competing event was, respectively, OTHER DEATHS (center: CHD as primary

AC

CE P

TE

D

MA

event) and CHD (right: OTHER DEATHS as primary event).

AC

CE P

TE

D

MA

NU

SC R

IP

T

ACCEPTED MANUSCRIPT

19

AC

CE P

TE

D

MA

NU

SC R

IP

T

ACCEPTED MANUSCRIPT

20

ACCEPTED MANUSCRIPT

21

HIGHLIGHTS

Cox and Fine-Gray models among 1677 middle aged men of a population study that

SC R



IP

T

Competing risks and lifetime coronary heart disease incidence during 50 years of follow-up

reached 50 years of follow-up were compared; We considered 10 selected risk factors;



Cholesterol was significantly and positively related to incidence of CHD contrasted with deaths from any other cause;

When the primary events were deaths from any other cause, cholesterol was inversely

MA



and age positively related;

Revisions of existing results whereby competing risks of CHD were not assessed may

CE P

TE

D

be needed.

AC



NU

