The clinical applicability of an automated plethysmographic determination of the ankle-brachial index after vascular surgery

Study population

A prospective cross-sectional study was performed between September 2013 and February 2014. Patients aged 18 years or older were eligible for inclusion in this study after they underwent a vascular intervention. Patients with preoperatively proven non-compressible arteries (ABI ≥ 1.4 using the standard manual method) or patients unable to consent were excluded from this study.

Data collection

Patient characteristics were acquired by questionnaires from the consent form and patient records. ABIs were obtained the first postoperative day by a manual and an automated plethsmographic method (Dopplex® ABIlity, Huntleigh diagnostics, Inc) on the surgical ward. All measures were performed two-sided (the right side first), first by the manual method followed by the automated method on subjects in supine position. To mimic the daily practice, the manual ABIs were obtained by trained operators and the automated ABIs by registered nurses. Time measures were obtained using a stopwatch.

Statistics

According to the protocol, 38 patients were needed to detect a correlation coefficient between the manual and automated method of 0.5 with 90% power using a test size of 0.05 (two-sided).

Time performance

Time management between both methods was tested using the paired t-test.

Clinical applicability: Agreement

The four repeated ABI-measurements (2 sides × 2 methods) were analyzed using linear mixed modeling. As fixed effects on ABI, the mean difference between both methods and the effect of operation of the leg were estimated. The following random effects as components of total variance of ABI were considered: (i) between-subjects, (ii) side nested within subject, and (iii) method nested within subject and side. The correlation coefficient between methods was calculated as the sum of the variance components (i) and (ii) and expressed as proportion of total variance (i) + (ii) + (iii). Limits of agreement ¹⁰ of the difference between both methods were calculated as the estimated fixed mean difference plus or minus 1.96 times the square root of twice the component (iii).

Clinical applicability: Precision

Linear mixed modeling was used to estimate the within-subject variance of the ABI from its observed left–right differences under either method and to test the null hypothesis that this variance was the same under both methods using a likelihood ratio test. This variance is built up from random measurement variability along with within-subject variability based on almost two simultaneous measurements within a subject. Of the former linear mixed model analysis, we took the variance component (ii) to subtract it from the variance obtained from this mixed model analysis, so that the remaining variance is supposed to be completely attributable to the method used and to stand for the intrinsic precision of the method. In order to better compensate for potential bias due to missing values of ABI-measurements, adjustment was made for the following fixed effects in this analysis: method (manual/automated), left leg operated, right leg operated, gender, age, Rutherford score >2, and current smoker.

Clinical applicability: ABI failure

For analyzing the probability of a measurement method not producing a valid ABI-value, a dichotomous failure indicator was defined. This failure outcome variable with four repeated measurements within a subject was analyzed through generalized linear modeling (GENLIN). The generalized estimating equations method was used to account for the correlation between the repeated measurements within a subject. GENLIN with an identity link function was used for testing the difference in failure rate between both measurement methods. Given the presence of a valid manual measurement, GENLIN was used with a logit link function for analyzing the automated failure rate depending on the following explanatory variables: gender, current smoker, side, if the leg was operated, age, and the value of the manual ABI measurement.

Statistical analyses were performed using the SPSS® software program version 20 (IBM, Armonk, NY, USA). P-values below 0.05 were considered to denote statistical significance.

Time performance

A total of 42 patients were included in this study (Table 1). The automated method took 9.3 min: 1.1 min shorter than the hand method (95% CI: −0.2 to +2.4; P = 0.095).

OPEN IN VIEWER

Table 1.

Patient characteristics (n = 42).

Characteristic			n (%)
Gender (male/female)
Male			31 (74)
Female			11 (26)
Risk factors
Age			69 ± 7.8
Current smoker			38 (90)
Peripheral arterial disease (Rutherford class)
≤2			13 (31)
≥3			29 (69)
Procedure
Central			15 (36)
Peripheral			27 (64)
Raw ABI data
Side	Method	n	Mean ± SD	Median (range)
Right	Manual	40	0.81 ± 0.28	0.82 (0.30–1.61)
	Automated	26	0.96 ± 0.26	0.96 (0.57–1.59)
Left	Manual	41	0.81 ± 0.26	0.87 (0.36–1.38)
	Automated	25	0.96 ± 0.29	0.86 (0.59–1.97)

Clinical applicability: Agreement

The automated method yielded a significantly higher mean ABI than the manual method (0.105, 95% CI: 0.017 to 0.193; P = 0.020). Operation was associated with a significantly lower mean ABI (0.092; 95% CI: 0.014 to 0.170; P = 0.022). After adjustment for these fixed effects, total variance was 0.0759 (ABI squared) and could be decomposed into: (i) 0.0302, (ii) 0.0156, and (iii) 0.0302. The correlation between both methods was 0.60. The limits of agreement of the difference (automated − manual) were −0.376 and +0.587 (Figure 1). OPEN IN VIEWER

Clinical applicability: Precision

The within-subject variance of ABI based on left–right differences of the automated method was 0.0186 compared with 0.0315 of the manual method. However, this difference was not significant (chi-square with 1 df was 3.758, P = 0.053). Assuming the within-subject variance to be the same under both methods its estimate was 0.0273. Subtracting from these variances the pure within-subject component (ii) of 0.0156, we are supposed to get the intrinsic measurement variance (precision) of the methods: 0.0159 for the manual method and 0.0030 for the automated method, or 0.0117 assuming them to be equal. The absolute difference between duplicate ABI-measurements using the manual method in a patient under exactly the same conditions was maximally 1.96 × √(2 × 0.0159) = 0.350 at the 95% level. For the automated-method, this maximum absolute difference was 0.152 and assuming identical variances of both methods was maximally 0.300.

Clinical applicability: ABI failure

The automated method had a failure rate of 39%: 34% points (95% CI: 18.7 to 49.5, P < 0.0005) than the manual method. The odds of the automated method not producing a valid ABI-value was significantly influenced by gender and the level of the manually-measured ABI-value. The failure odds ratio of females to males was 10.74 (95% CI: 1.78 to 65.0; P = 0.010). A 0.1 point decrease in the manually measured ABI-value was associated with a failure odds ratio of 1.24 (95% CI: 1.02 to 1.52; P = 0.035). These odds ratios were adjusted for the other variables in the model (current smoker, side, operation of the leg, and age).