Prediction rule for cardiovascular events and mortality in peripheral arterial disease patients: Data from the prospective Second Manifestations of ARTerial disease (SMART) cohort study

Article Outline

• Abstract
• Methods
  • Study population
  • Predictors
  • Outcome
  • Data analysis
• Results
  • Study population
  • Prediction model and model performance
  • Score chart
• Discussion
  • Generalizability
  • Score chart
  • Extensive workup
• Conclusions
• Author contributions
• Appendix
• References
• Copyright

Background

Patients with peripheral arterial disease (PAD) are at high risk of secondary cardiovascular death and events such as myocardial infarction or stroke. To minimize this elevated risk, cardiovascular risk factors should be treated in all PAD patients. Secondary risk management may benefit from a prediction tool to identify PAD patients at the highest risk who could be referred for an additional extensive workup. Stratifying PAD patients according to their risk of secondary events could aid in achieving optimal therapy compliance. To this end we developed a prediction model for secondary cardiovascular events in PAD patients.

Methods

The model was developed using data from 800 PAD patients who participated in the Second Manifestations of ARTerial disease (SMART) cohort study. From the baseline characteristics, 13 candidate predictors were selected for the model development. Missing values were imputed by means of single regression imputation. Continuous predictors were truncated and transformed where necessary, followed by model reduction by means of backward stepwise selection. To correct for over-fitting, a bootstrapping technique was applied. Finally, a score chart was created that divides patients in four risk categories that have been linked to the risk of a cardiovascular event during 1- and 5-year follow-up.

Results

During a mean follow-up of 4.7 years, 120 events occurred (27% nonfatal myocardial infarction, 21% nonfatal stroke, and 52% mortality from vascular causes), corresponding to a 1- and 5-year cumulative incidence of 3.1% and 13.2%, respectively. Important predictors for the secondary risk of a cardiovascular event are age, history of symptomatic cardiovascular disease, systolic blood pressure, high-density lipoprotein cholesterol, smoking behavior, ankle-brachial pressure index, and creatinine level. The risk of a cardiovascular event in a patient as predicted by the model was 0% to 10% and 1% to 28% for the four risk categories at 1- and 5-year follow-up, respectively. The discriminating capacity of the prediction model, indicated by the c statistic, was 0.76 (95% confidence interval, 0.71-0.80).

Conclusion

A prediction model can be used to predict secondary cardiovascular risk in PAD patients. We propose such a prediction model to allow for the identification of PAD patients at the highest risk of a cardiovascular event or cardiovascular death, which may be a viable tool in vascular secondary health care practice.

Peripheral arterial disease (PAD) is a substantial health care problem in Western societies. A prevalence of 3% to 10% has been reported for the general population,¹ and a prevalence up to 29% has been reported in primary health care populations.² These figures will likely increase in the coming years, due to aging and increased life expectancy,³ in part by the improved treatment of coronary artery disease (CAD), and the increasing prevalence of diabetes mellitus.

PAD is associated with high rates of cardiovascular events, including nonfatal myocardial infarction (MI) and stroke, and vascular death.², ⁴ Although PAD patients are exposed to the same risk factor profile for cardiovascular diseases as CAD patients and cerebrovascular disease (CVD) patients,⁵ the incidence of future cardiovascular events in PAD patients is higher than it is in CAD or CVD patients.⁶ This is probably because a large proportion of PAD patients also have established atherosclerotic disease in other vascular beds⁶ and because PAD symptoms have a relatively late onset.

Secondary risk management in PAD patients, which currently consists of treatment of cardiovascular risk factors and lifestyle interventions, is therefore of particular importance. Recent data show that appropriate risk factor control is reached less frequently in PAD patients than it is in CAD or CVD patients,⁷ stressing the need of further optimization of cardiovascular risk factor management in PAD patients. Secondary risk factor management in PAD may also benefit from a prediction model that could identify PAD patients that are at the highest risk of a cardiovascular event. These highest-risk PAD patients could then be monitored and treated even more intensively to prevent a secondary event. Stratifying PAD patients according to their risk of secondary events could also aid in achieving optimal cardiovascular risk factor management.

Primary risk assessment has mainly been based on traditional cardiovascular risk factors. However, a prediction model for the secondary risk in patients with overt PAD may benefit from additional information on comorbid diseases, the clinical staging of PAD, hemodynamic variables such as the ankle-brachial index (ABI), and biologic markers.⁸ No prediction model for the risk of cardiovascular events in patients with established PAD that incorporates these characteristics is currently available. The objective of this study was to develop such a prediction model to identify PAD patients at the highest risk of a cardiovascular event or cardiovascular death.

Back to Article Outline

Methods 

Study population 

This study used data from patients enrolled in the Second Manifestations of ARTerial disease (SMART) study. The SMART study is an ongoing, prospective, single-center cohort study in patients with clinically manifest vascular disease or cardiovascular risk factors.⁹ The main inclusion criteria are CAD, CVD, or PAD, abdominal aortic aneurysm (AAA), or any or all of the following risk factors for atherosclerosis: hyperlipidemia, diabetes mellitus (type 1 and 2), or hypertension. Excluded are patients with a terminal malignancy, patients not able to live independently (Rankin scale >3), or patients who are not sufficiently fluent in Dutch.

Since 1996, patients have been included who were aged 18 to 80 years and were referred to the University Medical Center (UMC) Utrecht, The Netherlands, for the treatment of clinically manifest vascular disease or cardiovascular risk factors. Patients underwent a vascular screening, including a questionnaire, blood chemistry analysis, and noninvasive laboratory testing, which included a 12-lead resting electrocardiogram (ECG), resting ABI, and ABI after a treadmill test. Ultrasound imaging was used to measure the juxtarenal and infrarenal anteroposterior diameter of the aorta and the length and volume of both kidneys. Hemodynamically significant stenosis of the common and internal carotid artery was assessed with color Doppler-assisted duplex scanning. Ultrasound imaging was used to measure common carotid intima-media thickness, and common carotid distensibility was documented with the Wall Track System (Pie Medical Systems, Maastricht, The Netherlands).

Eligible patients received written and oral information about the goals and methodology of the study from qualified research nurses or doctors at their first or second visit to the hospital, and all patients were asked to provide written informed consent. The study was approved by the UMC Utrecht Medical Ethics Committee. The rationale and design of the SMART study have been described in detail elsewhere.⁹

Recruitment in the SMART study is ongoing, and approximately 800 patients are included annually. For the current study, analyses were limited to patients included between September 1996 and March 2007 who were referred to the UMC Utrecht for clinically manifest PAD documented as resting ABI ≤0.90, postexercise ABI decreasing ≥20% in at least one leg, rest pain or gangrene/ulcers, with signs of intermittent claudication.⁹ The study included 800 patients with PAD.

Predictors 

All patients underwent a noninvasive standardized diagnostic protocol on a single day at the UMC Utrecht. Medical history, current and past smoking behavior and alcohol consumption were derived from a standardized health questionnaire described elsewhere.⁹ A physical examination included measurement of weight and height with the participants wearing indoor clothes and no shoes. Diastolic and systolic blood pressure was measured twice in most patients, while seated, in the right and left upper arm, and the mean value was taken as the blood pressure.

The right and left ABI at rest were determined in supine patients by measuring the blood pressure in the arm and the two pedal arteries at the ankle for each side. The ratios of the highest systolic blood pressure measured at the ankle to the highest systolic blood pressure measured in both arms were calculated for each leg. The ABI of the leg with the lowest ratio is reported as lowest ABI.

Blood samples were collected after overnight fasting. Levels were measured of total cholesterol, triglycerides, high-density lipoprotein (HDL) cholesterol, glucose, homocysteine, creatinine, and high-sensitive C-reactive protein (hsCRP). Low-density lipoprotein (LDL) cholesterol was calculated with the Friedewald formula when the triglyceride plasma level was <4.5 mmol/L. Diabetes was defined as a fasting plasma glucose ≥7.0 mmol/L, a nonfasting serum glucose ≥11.1 mmol/L, or the use of oral antidiabetic drugs or insulin.

Outcome 

Patients were biannually asked to complete a questionnaire on hospitalizations and outpatient clinic visits. The outcome of interest was the occurrence of a cardiovascular event comprising nonfatal MI, nonfatal stroke, or death from vascular causes.

Nonfatal MI was defined as an event with at least two of the following: (1) chest pain for ≥20 minutes not disappearing after administration of nitrates, (2) ST elevation >1 mm in two following leads or a left bundle branch block on the ECG, or (3) creatine kinase (CK) elevation of at least two times the normal value of CK and a MB-fraction >5% of the total CK value.

Nonfatal stroke was defined as relevant clinical features that caused an increase in handicap of at least one grade on the modified Rankin scale, accompanied by a fresh infarct or a hemorrhage on a repeated computed tomography (CT) scan, or as clinical deficits that caused an increase in handicap of at least one grade on the modified Rankin scale for which no CT documentation was needed.

Vascular death was any sudden death, including unexpected cardiac death occurring ≤1 hour after onset of symptoms or within 24 hours given convincing circumstantial evidence, or death from stroke, MI, congestive heart failure, or an AAA rupture.⁹ When a possible event was noted, hospital discharge letters and results of relevant laboratory and radiology examinations were collected. For each event, written standard operating procedures were followed to classify each event accordingly. Three members of the SMART Study Endpoint Committee, consisting of physicians from different departments, conducted an independent audit of this information. In case of disagreement, opinions of other members of the Endpoint Committee were sought and final adjudication was based on the majority of the classifications obtained. The first occurring cardiovascular event per patient was used as an outcome event in the prediction model.

Data analysis 

Baseline characteristics of the PAD patient population included in this study are presented in Table I. The median age was 58 years, and 66% of the patients were men. Diabetes was present in 17%. Most patients had intermittent claudication (Fontaine class 2), but only 8% had critical limb ischemia (Fontaine class 3 or 4). The PAD patients had a median body mass index (BMI) of 25.7 kg/m².

Table I. Baseline characteristics of 800 peripheral artery disease patients in the SMART study data

Variable	Missing values	No. (%), or median (IQR)
Age, y	0	58(51-68)
Gender, male	0	526(66)
History of symptomatic cardiovascular diseases	0	207(26)
Cerebrovascular disease	0	66(8)
Coronary artery disease	0	166(21)
Body mass index, kg/m2	2	25.7(23.4-28.3)
Diabetes	37	137(17)
Hypertension	12	418(52)
Systolic blood pressure, mm Hg	0	143(131-160)
Diastolic blood pressure, mm Hg	1	81(74-88)
Hyperlipidemia	15	554(69)
Total cholesterol, mmol/L	5	5.6(4.8-6.4)
HDL-cholesterol, mmol/L	5	1.16(0.94-1.41)
LDL-cholesterol, mmol/L	58	3.53(2.79-4.26)
Triglycerides, mmol/L	5	1.67(1.20-2.37)
Hyperhomocysteinemia	151	99(12)
Homocysteine, μmol/L	151	12.6(10.2-15.5)
Smoking behavior	9
Never smoked	...	74(9)
Former smoker	...	568(71)
Currently smoking	...	149(19)
Alcohol consumption	12
Never used alcohol	...	170(21)
Former user of alcohol	...	367(46)
Currently using alcohol	...	251(31)
Fontaine classification	0
Grade 2: intermittent claudication	...	737(92)
Grade 3/4: critical limb ischemia	...	63(8)
Ankle-brachial pressure index, lowest	16	0.69(0.55-0.85)
Creatinine, μmol/L	9	86(75-100)
High-sensitive C-reactive protein, mg/L	298	2.60(1.30-5.63)

HDL, High-density lipoprotein; IQR, interquartile range (25th to 75th percentile); LDL, low-density lipoprotein; SMART, Second Manifestations of ARTerial disease.

After a critical literature review, the 13 candidate predictors (Table II) were selected from these characteristics for the prediction model. Missing values of patient characteristics, comprising about 5% of all required values, were imputed as determined by the correlation between patient characteristics with missing values with the other variables by means of single regression imputation. Continuous candidate predictors were examined for extremes, and extreme values were truncated to the first and ninety-ninth centiles to prevent distortion of the relationship between the candidate predictor and the outcome.¹⁰ Furthermore, visual inspection of the relationship between continuous candidate predictors and the outcome was used to apply nonlinear transformations where necessary.

Table II. Hazard ratios and the contribution of selected predictors to the multivariate Cox regression model (χ² and df) in a full model for cardiovascular events, based on a single imputed data set with n = 800

Predictor	HR (95% CI)	χ2	df
Age >55 years	1.9(1.4-2.7)	14	1
Male gender	1.0(0.7-1.6)	0.01	1
History of symptomatic cardiovascular disease	3.0(1.6-5.5)	11	1
Body mass index	0.9(0.7-1.1)	2	1
Diabetes	1.3(0.8-2.0)	1	1
Systolic blood pressure	1.2(0.9-1.5)	2	1
High-density lipoprotein cholesterol	0.8(0.6-1.0)	3	1
Homocysteinea	1.2(0.9-1.6)	2	1
Current or former smoker	2.0(1.0-4.0)	4	1
Fontaine classification grade 3/4: CLI	1.2(0.6-2.1)	0.24	1
Ankle-brachial pressure index	0.7(0.6-1.0)	5	1
Creatinine	1.2(1.1-1.3)	10	1
High-sensitive C-reactive protein	1.1(1.0-1.2)	1	1

CI, Confidence interval; CLI, critical limb ischemia; df, degrees of freedom; HDL, high-density lipoprotein; HR, hazard ratio.

The presence of previous symptomatic CAD or CVD was summed in a single variable for previous cardiovascular disease, with a score of 0 for no history, 1 for the presence of either condition, and 2 if CAD and CVD were both present in the patient's history. Smoking behavior was redefined into two groups, consisting of patients who never smoked or of patients who currently smoke or had a history of smoking, before further analysis to abolish a possible reporting bias by patients. Next, reduction of the prediction model was performed by means of backward stepwise selection of predictors using the Akaike information criterion resulting in the final (reduced) model.¹¹

Any prediction model shows too optimistic performance—over-fitting—in the data set from which it has been developed.¹² This over-fitting can be corrected by applying a bootstrapping technique. With this technique, the modeling process is repeated multiple times to validate, and if necessary, adjust the regression coefficients of the prediction model with a shrinkage factor. The final model's predictive performance after bootstrapping can be regarded as the expected performance in similar future patients.

Model performance was examined by determination of the model's discrimination and calibration. Discrimination indicates how well the prediction model is able to distinguish between patients who will experience the outcome and those who will not. Discrimination was assessed by calculating the concordance (c) statistic with 95% confidence intervals (CIs). The interpretation of the c statistic is equivalent to the area under the receiver operating characteristic curve; that is, a c statistic of 0.5 indicates no discrimination above chance, whereas a c statistic of 1.0 indicates perfect discrimination. Model calibration—the agreement between predicted risks and observed risks—was assessed by comparing the predicted survival and the observed survival at the 1- and 5-year follow-up.

To facilitate the calculation of the individual risk, a score chart was created by dividing each regression coefficient by the smallest regression coefficient and rounding it to the nearest integer. The sum of scores of all predictors was used to classify the patient in one of four risk categories, each of which has been linked to the risk of a cardiovascular event during 1 and 5 years of follow-up.

All variables were handled in an SPSS 16.0 data sheet (SPSS Inc, Chicago, Ill). Analyses were performed using R 2.7.2 software (R Foundation for Statistical Computing, Vienna, Austria).

Back to Article Outline

Results 

Study population 

The analyses included 800 PAD patients. Baseline characteristics are presented in Table I. The mean follow-up was 4.7 years (range, 0-10 years). Full follow-up data were available for 95.9% of patients. At some time during the study, 33 patients (4.1%) were lost to follow-up, but partial data were available for these patients out to a mean follow-up of 1659 days (range, 215-3420 days). A total of 120 events occurred (27% nonfatal MI, 21% nonfatal stroke, and 52% mortality from vascular causes) corresponding to a 1- and 5-year cumulative incidence of 3.1% and 13.2%, respectively.

Prediction model and model performance 

According to the visual inspection of the relationship between age and outcome, no age effect was assumed for age ≤55 years, but a linear age effect was assumed for age >55 years. In addition, a logarithmic transformation was applied for homocysteine. All other selected continuous predictors were added to the model as linear variables. The full main effects model had a model χ² of 146 and an R² value of 0.154. Predictors with a large prognostic strength were age (χ², 14), a history of symptomatic cardiovascular diseases (χ², 11), and plasma levels of creatinine (χ², 10).

Backward stepwise model reduction resulted in sex, BMI, diabetes, homocysteine, Fontaine classification, and hsCRP being dropped from the model. After 100 repeats, the bootstrap procedure resulted in a shrinkage factor of 0.88, which was used to correct the regression coefficients for over-fitting. Predictors and corrected regression coefficients in the final model are presented in Table III. The R² value of the final model was 0.147.

Table III. Corrected regression coefficients of the predictors in the stepwise backward selected model

Predictor	Regression coefficient (95% CI)
Age >55 years	0.0489(0.0426to0.0690
History of symptomatic cardiovascular diseases, 0-2	0.5105(0.4281to0.7369)
Systolic blood pressure	0.0059(0.0026to0.0110)
High-density lipoprotein cholesterol	–0.4368(–0.7804to–0.2164)
Current or former smoker	0.6318(0.3788to1.0630)
Ankle-brachial pressure index	–0.9296(–1.5014to–0.6200)
Creatinine	0.0068(0.0063to0.0095)

CI, confidence interval.

The discriminating capacity of the final model, as indicated by the c statistic, was 0.76 (95% CI, 0.71-0.80). To provide an indication of the calibration of the model, Table IV reports the number of study patients across the four risk categories of the risk score and the observed number of events.

Table IV. Number of patients with and without an event during 1-year and 5-year follow-up across the categories of the risk score compared with the risk before the application of the model

Risk score	Total patients(n = 800) No. (%)	1-year FU Event (n = 24) No. (%)	5-year FU Event (n = 84) No. (%)
1. Low	91(11)	0(0)	1(1)
2. Moderate	340(43)	4(1)	17(5)
3. High	239(30)	7(3)	29(12)
4. Very high	130(16)	13(10)	37(28)
Risk without application of the prediction model		3.1%	13.2%

FU, Follow-up.

Score chart 

The Fig shows the score chart for predicting the risk of a cardiovascular event at 1 year and 5 years. After adding the scores for each predictor, the sum score will classify the patient in one of the four predicted risk categories. As an example of how to use this score chart, a man aged 60, with a systolic blood pressure of 135 mm Hg, a resting ABI of 0.83, an HDL level of 1.00 mmol/L, a creatinine level of 130 μmol/L, and a history of coronary artery disease, who quit smoking 5 years ago receives a score of: 40 (years >55 × 8) + 86 (CAD and no history of CVD) + 135 (systolic blood pressure) + 107 (former smoker) + 130 (creatinine level) – 74 (HDL × 74) – 130 (ABI × 157) = 294 points. This score places the patient in the third risk category with a 1-year and 5-year estimated risk of a cardiovascular event of 3% and 12%, respectively.

• 
  • View Large Image
  • Download to PowerPoint

• Fig. 

  Score chart for the 1-year and 5-year predicted risk of a cardiovascular event. The exact survival estimate can be calculated by S(t) = S0(t)^exp(LP), where the linear predictor (LP) is β₁ × x₁ + β₂ × x₂ + …, with x denoting the predictor and β the regression coefficient (Table III), and t is the time point of interest.

Table IV presents the number of study patients across the four risk categories of the risk score and the observed number of events. Respectively, zero (0%) and one patient (1%) in the low-risk group experienced an event after 1 and 5 years of follow-up, whereas 13 (10%) and 37 patients (28%) in the very-high risk group experienced an event after 1 and 5 years of follow-up. These risk categories can be used to identify patients at low and high risk.

Back to Article Outline

Discussion 

Patients with PAD experience a higher rate of cardiovascular events death caused by an atherothrombotic event, such as MI and stroke, than patients with other primary manifestations of atherosclerotic disease.⁶ The presence of several risk factors and comorbidity within the PAD group importantly contributes to this high risk. This warrants strict risk factor treatment of all PAD patients according to the current international guidelines. In this study we developed a prediction rule to identify those PAD patients that are at the highest risk of a cardiovascular event or death at 1 year and 5 years of follow-up.

From the traditional risk factors included in the model, male sex, obesity (BMI), and diabetes were dropped as predictors for a secondary cardiovascular event during the stepwise backward selection procedure. Studies show these risk factors contribute to the development of atherosclerosis and to the prediction of a primary cardiovascular event, but they do not contribute to the risk prediction of a secondary cardiovascular event in PAD patients in our model. A possible explanation for this could be that these risk factors are involved in the etiology of atherosclerosis but are less significant in its progression. In fact, male sex has been associated with better risk factor management in patients with established atherosclerotic disease,⁷ which in turn could slow disease progression. Furthermore, in a large, international study on the 1-year cardiovascular event rate in a stable outpatient population similar to the SMART cohort, correction for obesity (BMI) did not substantially affect this 1-year event rate.⁶

From the nonclassical risk factors for which an additional predictive effect has been suggested in the literature, blood homocysteine level, Fontaine classification, and hsCRP level were dropped from the final model. The association between homocysteine levels and the severity of atherosclerosis, progression of the disease, and mortality rates has been discussed in several retrospective and prospective studies,¹³ and folic acid and B-vitamin supplementation have been proposed as treatment for hyperhomocysteinemia. More recent studies, however, indicate that although such therapy significantly lowers homocysteine levels, it has no effect on vascular inflammation¹⁴ or the risk of a cardiovascular event.¹⁵ These observations are in line with our finding that homocysteine has no additive contribution to the prediction of a secondary cardiovascular event in PAD patients. We note that this concerns the predictive value of homocysteine levels at baseline, and that no conclusions can be drawn from this research about the effects of therapy to lower homocysteine on the progression of the disease.

Elevated CRP levels have also been associated with the occurrence of cardiovascular events and appear to contribute to the risk prediction of recurrent cardiovascular events in patients with coronary heart disease.¹⁶ Data are limited on the added value of CRP to the risk prediction of secondary cardiovascular events in PAD patients. One prospective study demonstrated that high levels of CRP were independently associated with future cardiovascular events in PAD patients.¹⁷ This contradicts our findings that CRP levels did not have a significant contribution to the prediction of secondary events in PAD patients. The difference might be explained by the fact that our end point does not include revascularization procedures, whereas coronary and lower extremity revascularizations comprised most of the events in the combined end point of this other study. Furthermore, a possible publication bias for negative results on the association between CRP levels and cardiovascular events might explain the still limited number of publications on this topic. On the other hand, an association of CRP levels with renal function has been reported.¹⁸, ¹⁹ The predictive value of CRP might therefore already be contained in the predictive value of the creatinine level. Creatinine, being the strongest predictor of both, was kept in the model during backward selection.

Clinical disease staging for the severity of PAD had additional value in secondary risk prediction. Clinicians use the Fontaine classification to designate patients as asymptomatic and symptomatic and to subdivide symptomatic patients into groups according to the severity of their complaints. The prognosis of patients with critical limb ischemia (Fontaine classification grade 3 and 4) is substantially worse than the prognosis of patients with intermittent claudication (Fontaine classification grade 2); therefore, the use of the Fontaine classification as a possible predictor for future cardiovascular events has been suggested. However, the Fontaine classification was dropped in the stepwise backward selection procedure during the development of our final model. A possible explanation for this is that the number of patients with critical limb ischemia in our cohort was rather small (8% of all patients). Such a limited number of cases in one group of a dichotomous variable may abolish its predictive effect.

Furthermore, the ABI, as another marker for the severity of PAD, provided additional predictive information in our model. The information provided by the Fontaine classification might contain similar information as the stronger predictor ABI. The finding that the ABI has a predictive value in our model is consistent with several cohort studies demonstrating a strong association between decreased ABI and an increased occurrence of fatal and nonfatal cardiovascular events.

Generalizability 

The prediction rule has been developed on a heterogeneous, white majority population of PAD patients referred from primary care to secondary health care and may therefore be a viable tool in any vascular secondary health care practice. The model was developed from data of patients with established PAD who participated in the SMART study after their referral to UMC Utrecht. The patient characteristics of our study population (Table I) are comparable with those of other PAD populations, and we have no reason to assume that the study population used for the development of our prediction model was a selective group of patients. Furthermore, the exclusion criteria for the SMART study only exclude patients with a terminal malignancy, an inability to live independently, or those who are insufficiently fluent in Dutch. It is therefore unlikely that the generalizability of our prediction model to other PAD populations is thereby hindered.

The number of missing values was limited (<5%) for most baseline characteristics (Table I). Missing values for homocysteine and CRP occurred because these variables were not routinely measured during the first years of the study. These data are therefore “missing completely at random.”²⁰ The relative high fraction of missing values for LDL is related to the use of the Friedewald formula, with which LDL can only be calculated when the triglyceride plasma level is <4.5 mmol/L. These missing LDL values are “missing at random.”²⁰ The missing values listed in Table I do therefore not hinder the generalizability or validity of our prediction model.

During the development of the model, however, the regression coefficients were shrunk to correct for over-fitting and the generalizability (external validity) has not been tested in other populations than the SMART cohort. Therefore, generalizability testing in other populations should be considered before implementation of the model in clinical practice.

Score chart 

The score chart was developed to provide vascular surgeons and vascular internists with an easy tool to identify newly referred PAD patients at a high risk of a fatal or nonfatal MI or stroke. The elevated risk of secondary cardiovascular events warrants strict treatment of cardiovascular risk factors in all PAD patients. The prediction rule we developed could serve as a useful tool to identify the highest-risk patients, who could subsequently be referred for an extensive workup to identify—and when required and possible to treat—high-grade stenosis in coronary and carotid arteries to prevent these adverse outcomes.

Although the prediction rule allows the calculation of the cardiovascular risk at an individual patient level, we linked the score to risk categories, which may facilitate the decision of whether to refer a patient for an extensive workup. We have labeled the four categories as low, moderate, high, and very high risk. If patients with a low or moderate score (≤225) would be categorized as being at a lower risk and therefore would be treated according the current guidelines, 54% of all PAD patients would not be referred for an extensive workup, yet a cardiovascular event would occur in four patients (0.5%) during the 1-year follow-up and in 18 (2.3%) by 5 years. With a score of 225 as a cutoff for referral, the prediction model has a sensitivity of 83% and 79% at 1 and 5 years and a specificity of 55% and 58%, respectively.

These figures imply that a rather large number of patients need to be referred for an extensive workup as a result of a high predicted risk of a cardiovascular event by the model without actually experiencing such event (positive predictive value of 5.4% and 17.9% for 1-year and 5-years of follow-up). More important, however, the number of patients not referred for an extensive workup who will actually experience a cardiovascular event is low, with negative predictive values of 99.1% and 95.8% for 1-year and 5-years of follow-up. Five to 15 patients need to be referred for an extensive workup to find one high-risk patient that should (if possible) be treated to prevent secondary cardiovascular disease.

To date, no prediction model for assessing the secondary risk of a cardiovascular event in PAD patients has been developed, and besides the preoperative screening of PAD patients eligible for surgery, screening, and referral of other PAD patients is generally not performed. Therefore, referral of the very-high-risk group only (cutoff score >350) could already reduce the number of occurring secondary cardiovascular events, while the number of patients referred for an extensive workup remains limited. When only the very-high-risk group is referred, 16% of all PAD patients will be referred for an extensive workup. In the group not referred for workup, 11 (1.6%) and 47 (7.0%) cardiovascular events will occur at the 1-year or 5-year follow-up, respectively. With a cutoff value of 350, the prediction model has a sensitivity at 1 year and 5 years of follow-up of 54% and 44% and a specificity of 85% and 87%, respectively. When only the very-high-risk group is referred, two patients need to be referred for an extensive workup to find one high-risk patient who should be treated to prevent a secondary cardiovascular disease.

Despite the number of events being too limited to allow for accurate long-term risk prediction, the chart of the Fig clearly illustrates the fate of PAD patients across the four risk categories for follow-up times >5 years. The number of event-free patients categorized by the model as being at low risk (categories 1 and 2) visibly secedes from the number of event-free patients who were categorized as being at high risk (categories 3 and 4) at 10 years of follow-up. These trends underline the need for risk stratification for cardiovascular events in PAD patients.

Extensive workup 

The extensive workup we mention in this article should ideally be aimed at identifying the presence of pathology that may result in a cardiovascular event, such as a high-grade stenosis in the coronary or carotid arteries or cardiac arrhythmia. The discussion about what imaging modality, for example, carotid ultrasound scans, stress test, or angiography, should be used for these examinations might be strenuous but is beyond the scope of this article. We note, however, that a possible increase in morbidity and death resulting from interventions performed as part of an extensive workup are currently not incorporated into this prediction rule.

Nevertheless, further optimization of the treatment of PAD patients and, in particular, the treatment of patients at the highest risk remains advantageous and might benefit from taking secondary risk stratification into account. Exemplary, setting even more strict targets for risk factor reduction (lower lipid plasma levels, lower blood pressure, steady blood glucose levels, absolute smoking cessation, more overweight reduction) for the highest-risk patients, for instance, might already prove to be an effective “extensive workup.” Moreover, stratifying PAD patients according to their risk of a secondary event could aid in achieving improved therapy compliance.

Back to Article Outline

Conclusions 

We developed a prediction model for patients with established PAD to identify PAD patients at the highest risk of a cardiovascular event or cardiovascular death. The model has a high sensitivity for risk stratification at 1 and 5 years. Important predictors for the secondary risk of a cardiovascular event are age, history of symptomatic cardiovascular disease, systolic blood pressure, HDL cholesterol, smoking behavior, ABI, and creatinine. We believe this is the first prediction model for future cardiovascular events in patients with established PAD and that it may be a viable tool in vascular secondary health care practice.

Back to Article Outline

Author contributions 

Back to Article Outline

Appendix 

Members of the SMART Study Group are Ale Algra, MD, PhD, Pieter A. Doevendans, MD, PhD, Yolanda van der Graaf, MD, PhD, Diederick E. Grobbee, MD, PhD, L. Jaap Kappelle, MD, PhD, Willem P. Th. M. Mali, MD, PhD, Frans L. Moll, MD, PhD, Guy E.H.M. Rutten, MD, PhD, and Frank L. J. Visseren, MD, PhD.

Back to Article Outline

References 

1. Norgren L, Hiatt WR, Dormandy JA, Nehler MR, Harris KA, Fowkes FG, et al. Inter-society consensus for the management of peripheral arterial disease (TASC II). Eur J Vasc Endovasc Surg. 2007;33(suppl 1):S1–S75
   • View In Article
   • Full Text
   • Full-Text PDF (1906 KB)
   • CrossRef
2. Hirsch AT, Criqui MH, Treat-Jacobson D, Regensteiner JG, Creager MA, Olin JW, et al. Peripheral arterial disease detection, awareness, and treatment in primary care. JAMA. 2001;286:1317–1324
   • View In Article
   • MEDLINE
   • CrossRef
3. Hankey GJ. Vascular disease of the heart, brain and limbs: new insights into a looming epidemic. Lancet. 2005;366:1753–1754
   • View In Article
   • Full Text
   • Full-Text PDF (41 KB)
   • CrossRef
4. Hirsch AT, Halverson SL, Treat-Jacobson D, Hotvedt PS, Lunzer MM, Krook S, et al. The Minnesota regional peripheral arterial disease screening program: toward a definition of community standards of care. Vasc Med. 2001;6:87–96
   • View In Article
   • MEDLINE
   • CrossRef
5. Bhatt DL, Steg PG, Ohman EM, Hirsch AT, Ikeda Y, Mas JL, et al. International prevalence, recognition, and treatment of cardiovascular risk factors in outpatients with atherothrombosis. JAMA. 2006;295:180–189
   • View In Article
   • CrossRef
6. Steg PG, Bhatt DL, Wilson PW, D'Agostino R, Ohman EM, Rother J, et al. One-year cardiovascular event rates in outpatients with atherothrombosis. JAMA. 2007;297:1197–1206
   • View In Article
   • CrossRef
7. Cacoub PP, Abola MT, Baumgartner I, Bhatt DL, Creager MA, Liau CS, et al. Cardiovascular risk factor control and outcomes in peripheral artery disease patients in the Reduction of Atherothrombosis for Continued Health (REACH) Registry. Atherosclerosis. 2009;204:e86–e92
   • View In Article
   • Abstract
   • Full Text
   • Full-Text PDF (258 KB)
   • CrossRef
8. Haugen S, Casserly IP, Regensteiner JG, Hiatt WR. Risk Assessment in the patient with established peripheral arterial disease. Vasc Med. 2007;12:343–350
   • View In Article
   • CrossRef
9. Simons PC, Algra A, van de Laak MF, Grobbee DE, van der GY. Second Manifestations of ARTerial Disease (SMART) study: rationale and design. Eur J Epidemiol. 1999;15:773–781
   • View In Article
   • MEDLINE
   • CrossRef
10. McHugh GS, Butcher I, Steyerberg EW, Lu J, Mushkudiani N, Marmarou A, et al. Statistical approaches to the univariate prognostic analysis of the IMPACT database on traumatic brain injury. J Neurotrauma. 2007;24:251–258
    • View In Article
    • MEDLINE
    • CrossRef
11. Akaike H. A new look at the statistical model identification. IEEE Transactions on Automatic Control. 1974;19:716–723
    • View In Article
12. Harrell FE, Lee KL, Mark DB. Multivariable prognostic models: issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387
    • View In Article
    • MEDLINE
    • CrossRef
13. de Jong S. Hyperhomocysteinaemia in patients with peripheral arterial occlusive disease. Clin Chem Lab Med. 2001;39:714–716
    • View In Article
    • MEDLINE
    • CrossRef
14. Schernthaner GH, Plank C, Minar E, Bieglmayer C, Koppensteiner R, Schernthaner G. No effect of homocysteine-lowering therapy on vascular inflammation and haemostasis in peripheral arterial occlusive disease. Eur J Clin Invest. 2006;36:333–339
    • View In Article
    • MEDLINE
    • CrossRef
15. Albert CM, Cook NR, Gaziano JM, Zaharris E, MacFadyen J, Danielson E, et al. Effect of folic acid and B vitamins on risk of cardiovascular events and total mortality among women at high risk for cardiovascular disease: a randomized trial. JAMA. 2008;299:2027–2036
    • View In Article
    • CrossRef
16. Sabatine MS, Morrow DA, Jablonski KA, Rice MM, Warnica JW, Domanski MJ, et al. Prognostic significance of the centers for disease control/american heart association high-sensitivity C-reactive protein cut points for cardiovascular and other outcomes in patients with stable coronary artery disease. Circulation. 2007;115:1528–1536
    • View In Article
    • CrossRef
17. Beckman JA, Preis O, Ridker PM, Gerhard-Herman M. Comparison of usefulness of inflammatory markers in patients with versus without peripheral arterial disease in predicting adverse cardiovascular outcomes (myocardial infarction, stroke, and death). Am J Cardiol. 2005;96:1374–1378
    • View In Article
    • Abstract
    • Full Text
    • Full-Text PDF (136 KB)
    • CrossRef
18. Stuveling EM, Hillege HL, Bakker SJ, Gans RO, de Jong PE, de ZD. C-reactive protein is associated with renal function abnormalities in a non-diabetic population. Kidney Int. 2003;63:654–661
    • View In Article
    • MEDLINE
    • CrossRef
19. Verhave JC, Hillege HL, Burgerhof JG, Gansevoort RT, de ZD, de Jong PE. The association between atherosclerotic risk factors and renal function in the general population. Kidney Int. 2005;67:1967–1973
    • View In Article
    • MEDLINE
    • CrossRef
20. Greenland S, Finkle WD. A critical look at methods for handling missing covariates in epidemiologic regression analyses. Am J Epidemiol. 1995;142:1255–1264
    • View In Article
    • MEDLINE