Use of ankle brachial pressure index to predict cardiovascular events and death: a cohort studyBMJ 1996; 313 doi: http://dx.doi.org/10.1136/bmj.313.7070.1440 (Published 07 December 1996) Cite this as: BMJ 1996;313:1440
- G C Leng, clinical research fellowa,
- F G R Fowkes, professor of epidemiologya,
- A J Lee, research statisticiana,
- J Dunbar, research sistera,
- E Housley, consultant physicianb,
- C V Ruckley, professor of vascular surgeryb
- a Wolfson Unit for Prevention of Peripheral Vascular Diseases, Department of Public Health Sciences, University of Edinburgh, Edinburgh EH8 9AG
- b Royal Infirmary of Edinburgh NHS Trust, Edinburgh EH3 9YW
- Correspondence to: Professor Fowkes.
- Accepted 28 October 1996
Objective: To determine whether a low ankle brachial pressure index is associated with an increased risk of cardiovascular events and death, and whether the prediction of such events could be improved by including this index.
Design: Cohort study.
Setting: 11 practices in Edinburgh, Scotland.
Subjects: 1592 men and women aged 55–74 years selected at random from the age-sex registers of 11 general practices and followed up for 5 years.
Main outcome measures: Incidence of fatal and non-fatal cardiovascular events and all cause mortality.
Results: At baseline 90 (5.7%) of subjects had an ankle brachial pressure index </=0.7, 288 (18.2%) had an index </=0.9, and 566 (35.6%) </=1.0. After five years subjects with an index </=0.9 at baseline had an increased risk of non-fatal myocardial infarction (relative risk 1.38, 95% confidence interval 0.88 to 2.16), stroke (1.98, 1.05 to 3.77), cardiovascular death (1.85, 1.15 to 2.97), and all cause mortality (1.58, 1.14 to 2.18) after adjustment for age, sex, coronary disease, and diabetes at baseline. The ability to predict subsequent events was greatly increased by combining the index with other risk factors—for example, hypertensive smokers with normal cholesterol concentrations had a positive predictive value of 25.0%, increasing to 43.8% in subjects with a low index and decreasing to 15.6% in those with a normal index.
Conclusion: The ankle brachial pressure index is a good predictor of subsequent cardiovascular events, and improves on predictions by conventional risk factors alone. It is simple and accurate and could be included in routine screening of cardiovascular status.
In this study individuals with a low ankle brachial pressure index had an increased risk of fatal and non-fatal cardiovascular events
The index was a good predictor of subsequent cardiovascular events, and improved that of conventional risk factors alone
The ankle brachial pressure index could be included in routine screening of cardiovascular status
Individuals with a low ankle brachial pressure index require additional monitoring, and might benefit from aspirin or other secondary preventive measures
Coronary heart disease is the main cause of death and disability in elderly people,1 and numerous primary and secondary prevention trials have attempted to reduce its impact.2 Important risk factors include hypercholesterolaemia, hypertension, and cigarette smoking,3 and attempts have been made to target those at greatest risk by using scoring systems such as the Dundee method4 or by identifying subjects with early asymptomatic atheroma.5 Several cohort studies have shown that subclinical atherosclerosis is associated with an increased risk of subsequent cardiovascular events,5 6 7 8 but there is currently no universally accepted method for detecting early atheroma in the general population.
One commonly used non-invasive test of cardiovascular disease is the ankle brachial pressure index. It is quick and easy to measure, has high patient acceptability, and is an accurate and reliable indicator of atherosclerosis.9 10 11 Increased mortality has been shown to be associated with a low ankle brachial pressure index in patients referred for vascular investigation in Pittsburgh,12 in asymptomatic working men in Belgium7 and in elderly white women at risk of fracture.13 The ankle brachial pressure index has also been shown to predict cardiovascular events in conjunction with other tests,5 6 and as an independent measure,7 8 12 13 but before it can be considered as a possible screening tool its ability to predict cardiovascular events must be compared with that of conventional risk factors. We investigated the predictive value of the index in subjects included in the Edinburgh artery study.14
Subjects and methods
The Edinburgh artery study began in 1988 as a cross sectional survey of 1592 men and women aged 55–74 years. This population was selected at random, in five year age bands, from 11 general practices serving a range of socioeconomic and geographical areas throughout the city. The response rate was 65%, and follow up of a sample of non-responders showed no substantial bias. Details of recruitment have been described.14 Participants were followed up over five years to detect cardiovascular events and cause of death. At the end of five years, 102 subjects (6.4%) were lost to follow up. The study was approved by Lothian Health Board ethics subcommittee, and informed consent was obtained from all participants.
Subjects were invited to a university clinic where we administered a questionnaire including validated questions on smoking, history of diabetes, and angina from the World Health Organisation questionnaire.15 A clinical examination was then conducted by two pairs of specially trained nurses. They recorded systolic and diastolic (phase V) blood pressures in the right arm after 10 minutes' rest with a random zero sphygmomanometer. Ankle systolic pressures were measured in the posterior tibial artery of the right then left leg with a Doppler ultrasound probe (Sonicaid, Chichester) and a random zero sphygmomanometer with the cuff positioned proximal to the malleoli. The pulse was located with the Doppler probe, and the cuff inflated until the pulse was obliterated; the cuff was then deflated and the pressure recorded at the point when the pulse reappeared. A 12 lead electrocardiogram was recorded and coded independently by two observers using the Minnesota code.16
A sample of fasting blood was taken after five minutes' rest in the supine position to measure serum lipid concentrations, including total cholesterol, on a Cobas Bio analyser (Roche, Welwyn Garden City) with standard kits. Laboratory standardisation was carried out with commercially available standards (Wellcome scheme), and quality was assessed by examining systematic and random error against two control materials (Precipath universal bovine serum, Boehringer Mannheim, Lewes and pooled donated sera). The laboratory is standardised against the WHO Regional Lipid Reference Laboratory, Prague, Czech Republic. Fasting blood glucose concentration was also measured. Each subject was then asked to consume 75 g of glucose in a drink. A second blood glucose specimen was taken two hours after the oral glucose load.
IDENTIFICATION OF CARDIOVASCULAR EVENTS
We obtained information on cardiovascular events and deaths from general practitioners, who were asked to attach a card to the front of subjects' records at the start of the study and return it after a cardiovascular event or if the patient changed address or doctor; the information and statistics division of the Scottish Office Home and Health Department, which provided annual computer printouts of all hospital discharges in Scotland; hospitals, where medical records were investigated for discharges with relevant ICD-9 codes; and the participants themselves, who received an annual questionnaire asking about the development of chest pain, heart attack, and stroke as well as hospital attendances and general practitioner visits in the previous year.
To identify all deaths occurring in the cohort each participant's record was flagged at the NHS central registry so that certificates would be automatically forwarded. All deaths from cardiovascular causes were further investigated by using hospital or general practitioner records to ensure that the protocol criteria were fulfilled.
The criteria to define fatal or non-fatal myocardial infarctions and stroke were adapted from those proposed by the American Heart Association17 and are given in the appendix.
Information on the questionnaire and recording forms was checked by the clinic staff, coded, and entered on to a DBASE IV database. Error rates were determined by dual entry of all data, and any discrepancies were checked by reference to the original records.
The ankle brachial pressure index for each leg was calculated by dividing the ankle systolic pressure by the brachial systolic pressure. The lower of the indices obtained for the two legs was used as the measure of disease severity in the analysis.
Multiple events of the same type occurring in the same subject, such as two myocardial infarctions, were counted only once. The 2 test for trend was used to assess differences in non-fatal events and death between categories of subjects determined by baseline ankle brachial pressure index. Survival analysis based on the product limit or Kaplan-Meier estimate of the survival function was used to produce survival curves for non-fatal cardiovascular events and all cause mortality. The trend version of the Wilcoxon (Breslow) test was used to identify any survival differences after adjustment for the confounding effects of age and sex with a stratified analysis. Relative risks of fatal and non-fatal events were calculated for subjects with an ankle brachial pressure index at baseline </=0.9 compared with those with an index above 0.9. Relative risks were adjusted for age, sex, and presence of angina, myocardial infarction, or diabetes mellitus at baseline by multiple logistic regression.
To determine the usefulness of the index as a diagnostic test we calculated sensitivities, specificities, and likelihood ratios. We used a cut off point of 0.9 to define a low index in our calculations because it is a sensitive and specific measure of peripheral vascular disease in a clinical setting.9
All but two of the subjects were white, reflecting the composition of the general population of this age in Edinburgh. The distribution of the ankle brachial pressure index at baseline was slightly negatively skewed, with a mean of 1.03 (SD 0.18). Ten subjects refused to undergo the test, and of the remaining 1582 subjects, 90 (5.7%) had an index </=0.7, 288 (18.2%) </=0.9, and 566 (35.8%) </=1.0. One hundred and forty four subjects (9.0%) had a myocardial infarction during the five year follow up, of whom 55 died. A stroke occurred in 50 (3.1%) subjects, and 22 were fatal. Out of a total of 203 deaths, 89 (44%) were due to cardiovascular causes.
The lower the ankle brachial pressure index at baseline, the greater the occurrence of non-fatal myocardial infarction (P = 0.057) and stroke (P = 0.020) (table 1). A similar relation was seen between the index and death from myocardial infarction (P </= 0.001) and all cardiovascular causes (P </= 0.001). Deaths from non-cardiovascular causes or stroke were not significantly related to the index at baseline.
Figure 1 shows that the lower the ankle brachial pressure index at baseline, the lower the probability of survival (P</=0.001), independent of age and sex. Similar patterns were found for the probability of surviving without a non-fatal myocardial infarction or stroke. Both the probability of survival and of not having a myocardial infarction or stroke declined most when the index was below 0.9.
Patients with a baseline index </=0.9 had a slightly increased risk of non-fatal myocardial infarction (P = 0.085) and an increased risk of non-fatal stroke (P</=0.05), independent of age and sex (table 2). A low index was associated with increased relative risks of death from myocardial infarction (P</=0.01), all cardiovascular causes (P</=0.001), and all causes (P</=0.001), independent of age and sex. There was no increased risk of non-cardiovascular death. Adjustment for angina, myocardial infarction, or diabetes mellitus at baseline only slightly reduced the risks.
An ankle brachial pressure index </=0.9 showed moderate specificity (83.2%), low sensitivity (29.6%), and a likelihood ratio of 1.76 in predicting fatal and non-fatal cardiovascular events after five years (table 3). However, a lower index (</=0.7) showed better specificity (95.4%) and a higher likelihood ratio (3.07) but a lower sensitivity (14.2%). When subjects with diabetes were excluded the likelihood ratios were reduced (1.49 for an index </=0.9 and 1.52 for an index </=0.7).
The positive predictive value for a future cardiovascular event was 17.6% (95% confidence interval 13.1% to 22.1%) for subjects with an ankle brachial pressure index </=0.9 compared with 9.6% (8.0% to 11.2%) for those with an index >0.9. Positive predictive values were also higher for smokers (13.2%, 10.2% to 16.2%) than non-smokers (10.0%, 8.2% to 11.8%) and for hypertensive patients (15.0%, 11.7% to 18.3%) than those with normal blood pressure (9.4%, 7.7% to 11.1%), but the predictive value was lower for those with high cholesterol concentrations (10.5%, 8.6% to 12.4%) compared with those with normal values (12.0%, 9.3% to 14.7%). However, the ability to predict subsequent events was greatly increased by combining a low ankle brachial pressure index </=0.9 with other risk factors (table 4). In general, predictive values increased in the presence of a low index, particularly in normotensive non-smokers, and in smokers with low cholesterol concentrations. In most cases the predictive values for those with normal ankle brachial pressure indices were similar to those produced by risk factor status alone, with the exception of hypertensive smokers with a low cholesterol concentration, where the predictive value dropped from 25.0% to 15.6%.
Over 18% of subjects in this study had an ankle brachial pressure index </=0.9 at baseline. Thus almost one in five subjects would be identified as at risk should a population of this age be screened for early atherosclerosis. Other studies, however, have shown differing prevalences. In the cardiovascular health study only 12.4% of the population had an ankle brachial pressure index below 0.9,18 and in 40–55 year old men in Belgium only 3.8% had a low index.7 These differences undoubtedly reflect variations in the age structure of the study populations and the techniques of measurement as well as the underlying occurrence of atherosclerotic disease.
A relation between low ankle brachial pressure index and subsequent cardiovascular events might have been expected as lower limb disease is known to coexist with coronary and cerebrovascular disease.19 20 Indeed, previous studies in selected groups of subjects have all shown that a low ankle brachial pressure index is associated with reduced survival.7 8 9 10 11 12 13 This study shows that a low index is also associated with an increased risk of subsequent non-fatal cardiovascular events as well as death in the general population, independent of age, sex, and the presence of angina, myocardial infarction, and diabetes at baseline.
In this study, as in others, myocardial infarction was more common than stroke in subjects with lower limb disease.12 13 14 15 16 17 18 19 20 21 One Swedish study also showed an increase in non-cardiovascular mortality in those with a low ankle brachial pressure index, possibly because this population contained a relatively high proportion of smokers dying from other smoking related diseases such as neoplasm.22
WHAT SHOULD THE CUT OFF VALUE BE?
Most studies,7 8 9 10 11 12 13 including ours, show a significantly increased risk of death for an ankle brachial pressure index of 0.9 or lower. However, in the Pittsburgh study risk of a cardiovascular event became significant only for an index </=0.7,12 possibly because the overall prevalence of peripheral arterial disease in this sample was high, with over 75% of subjects having an index </=0.9. In our study the likelihood ratio of a subsequent event was greater for subjects with an index </=0.7 than those with an index </=0.9, reflecting the more advanced atheroma in those with a lower score. For the purposes of screening, however, an index </=0.9 would be more likely to identify those with early disease without symptoms, who would not otherwise seek medical attention.
An index of </=0.9 had a significantly greater predictive value for a subsequent cardiovascular event than higher values. Similarly, subjects who had hypertension or who smoked had higher predictive values than those who did not, but these differences were non-significant and lower than the predictive values associated with a low index. A low index reflects the combined effect of many risk factors over time and, once atherosclerosis has developed, would be expected to be a better predictor than any one risk factor alone. However, hypercholesterolaemia was associated with a lower predictive value than normal serum cholesterol concentration, possibly because in Edinburgh almost 65% of the baseline population had a cholesterol concentration >/=6.5 mmol/l. Alternatively, the predictive values of both high cholesterol concentration and hypertension may have been attenuated as a result of treatment given by general practitioners after the baseline examination. Total cholesterol is a poorer predictor than low density lipoprotein cholesterol,23 and although the predictive value could have been slightly improved by analysing low density lipoprotein, we used total cholesterol because it is the most widely measured parameter in general practice.
In an attempt to increase the accuracy of prediction of cardiovascular events from risk factors we included baseline ankle brachial pressure index in the calculation of positive predictive values. As expected, predictive values were generally higher in those with low ankle brachial pressure indices than in those with higher values, with the biggest effect occurring in those with a low cholesterol concentration. It is difficult to draw firm conclusions about the types of subjects in whom the index adds the most in terms of prediction, but it is possible that it will be most useful in subjects with few other risk factors who would otherwise be considered at low risk of an event.
A low ankle brachial pressure index is related to an increased risk of both fatal and non-fatal cardiovascular events. Since the measurement is simple and quick it could be carried out in general practice when screening for cardiovascular disease. In addition to the routine measurement of brachial blood pressure, systolic ankle pressures can be quickly measured with an ultrasound probe, which general practitioners already often use for obstetric purposes. It is probably unnecessary to measure the index in younger people, in whom atherosclerosis is unlikely to be present, but in middle aged and elderly subjects it should be measured when blood pressure, serum cholesterol, and smoking are assessed. A low index would indicate those subjects who require additional monitoring. In the absence of other cardiovascular risk factors, a low index might indicate a high risk individual who could benefit from aspirin or other secondary preventive measures. Before such preventive therapies are routinely recommended, however, it is essential that randomised controlled trials are performed to show their effectiveness in this group of high risk individuals.
We thank the following practices for their help: Bruntsfield Health Centre; S Channell and partners; J J C Cormack and partners; Crewe Road Medical Centre; Edinburgh University Department of General Practice; A Horne and partners; J G Ledingham and partners; Muirhouse Medical Group; I R F Ross and partners; Whinpark Medical Centre; and P White and partners. We also thank the clinic staff M Carson and A M Clark; the research secretary K Purves; M Whiteman and N Wright for data preparation; R J Prescott for statistical advice; and R A Riemersma for analysis of lipid concentrations.
Funding British Heart Foundation.
Conflict of interest None.