A new efficient trial design for assessing reliability of ankle-brachial index measures by three different observer groups

Abstract

Abstract Background The usual method of assessing the variability of a measure such as the ankle brachial index (ABI) as a function of different observer groups is to obtain repeated measurements. Because the number of possible observer-subject combinations is impractically large, only a few small studies on inter- and intraobserver variability of ABI measures have been carried out to date. The present study proposes a new and efficient study design. This paper describes the study methodology. Methods Using a partially balanced incomplete block design, six angiologists, six primary-care physicians and six trained medical office assistants performed two ABI measurements each on six individuals from a group of 36 unselected subjects aged 65–70 years. Each test subject is measured by one observer from each of the three observer groups, and each observer measures exactly six of the 36 subjects in the group. Each possible combination of two observers occurs exactly once per patient and is not repeated on a second subject. The study involved four groups of 36 subjects (144), plus standbys. Results The 192 volunteers present at the study day were similar in terms of demographic characteristics and vascular risk factors: mean age 68.6 ± 1.7; mean BMI 29.1 ± 4.6; mean waist-hip ratio 0.92 ± 0.09; active smokers 12%; hypertension 60.9%; hypercholesterolemia 53.4%; diabetic 17.2%. A complete set of ABI measurements (three observers performing two Doppler measurements each) was obtained from 108 subjects. From all other subjects at least one ABI measurement was obtained. The mean ABI was 1.08 (± 0.13), 15 (7.9%) volunteers had an ABI 1.4, i.e. a ratio that may be associated with increased stiffening of the arterial walls. Conclusion This is the first large-scale study investigating the components of variability and thus reliability in ABI measurements. The advantage of the new study design introduced here is that only one sixth of the number of theoretically possible measurements is required to obtain information about measurement errors. Bland-Altman plots show that there are only small differences and no systematic bias between the observers from three occupational groups with different training backgrounds.