Limited agreement between two noninvasive measurements of blood volume during fluid removal: ultrasound of inferior vena cava and finger-clip spectrophotometry of hemoglobin concentration

Approval for this study was obtained from the Institutional Review Board for Dartmouth College and Dartmouth Hitchcock Medical Center. We conducted an observational, cross-sectional study during 6 months in 2015 in two separate locations in Lebanon, New Hampshire: the inpatient unit at Dartmouth-Hitchcock Medical Center and the outpatient unit of Fresenius Kidney Care. Study investigators identified a convenience sample of spontaneously breathing adult patients who could comfortably recline. All patients had end-stage renal disease and required thrice-weekly hemodialysis. After written informed consent was obtained, investigators attempted an IVC ultrasound; participants with severe discomfort from the ultrasound probe, participants with severe ascites, and participants in whom the IVC could not be visualized were excluded.

Serial IVC ultrasounds and spectrophotometry of capillary hemoglobin concentration measurements were acquired during a single 4 h hemodialysis session. Hemodialysis protocols were not altered by participation, and standard bicarbonate dialysate and high-flux semisynthetic membranes were used. The initial ultrafiltration rate was set as the pre-dialysis weight minus the established target (or 'dry') weight divided by the planned duration (usually 4 h). If the target weight had not been established or was not known, the difference between the pre-dialysis weight and the antecedent post-dialysis weight (the interdialytic fluid gain) was used. Ultrafiltration rates were often adjusted to accommodate acute illnesses or when symptoms or signs arose during treatment that were attributable to fluid removal, such as nausea, cramping, or hypotension. The hemodialysis machines continuously measured the volume of fluid removed (ultrafiltration volume), which was compared against measures of blood volume change that were our research focus: maximum diameter of the inferior vena cava by ultrasound (${\rm IV}{{{\rm C}}_{{\rm US}}}$) and change in relative blood volume by capillary hemoglobin concentration from finger-clip spectrophotometry (RBV_SpHb).

One study investigator acquired all IVC ultrasounds using the same M-Turbo^™ ultrasound (Fujifilm Sonosite, Inc., Bothwell, Washington) equipped with a 1 to 5 MHz phased-array transducer. This investigator was a research coordinator whose central vein ultrasound acquisition skills had been verified (Lucas et al 2017). The investigator asked each participant to limit his or her movements and ensured that gurneys or chairs were kept at the same incline unless otherwise requested by dialysis staff. While each participant breathed regularly, the investigator acquired 10 s gray-scale (B-mode) video sequences along the longitudinal axis of the IVC from the subcostal window.

Ultrasounds were randomly arranged and interpreted by a second study investigator, who was blinded to both participant information and the order in which ultrasounds were acquired. The second investigator, a board-certified general internist who regularly used central vein point-of-care ultrasound since completing a training program 10 years ago (Lucas et al 2009), measured the IVC luminal diameters 1–2 cm distal from the hepatic vein inlet or, equivalently, 3–4 cm distal from the junction of the right atrium; this site was chosen because it is more compliant and, therefore, responsive to blood volume changes than more proximal sites (Wallace et al 2010). Although more than a half-dozen types of static and dynamic measurements of the IVC have been proposed, we focused on the maximum diameter because of simplicity (Moreno et al 1984), accuracy (Brennan et al 2007), and reliability (Fields et al 2011). To help gauge each participant's baseline status, we calculated a baseline IVC collapsibility index as the difference between the maximum and minimum IVC diameters (corresponding to end-expiration and end-inspiration, respectively) divided by the maximum diameter (Rudski et al 2010).

Spectrophotometry of capillary hemoglobin concentrations (SpHb) were obtained with the Radical-7 Pulse CO-Oximeter equipped with Rainbow^® R1-25L disposable adhesive sensors (Masimo Corporation, Irvine, CA). The Radical-7 continuously displays the perfusion index, which is the ratio of pulsatile to non-pulsatile blood flow. A perfusion index of 1.0 or lower indicates low pulse strength and signal quality. Low readings prompted investigators to ensure that the sensor was attached properly and that the site of the sensor was warmed by placing it under a blanket. If a low perfusion index persisted, the sensor was moved to a better perfused site (a different finger). Sensors always remained contralateral to arteriovenous fistulas, however, to avoid possible interference (Lin et al 1997). Because sphygmomanometers were also placed contralateral to fistulas as part of routine protocol, and because sphygmomanometers reduce pulse strength when inflated, the perfusion index was unavoidably low during blood pressure readings in patients with arteriovenous fistulas. Nevertheless, such measurements comprised less than 5% of SpHb readings and were easily discarded along with other SpHb measurements obtained when the perfusion index was low.

Measurements began with the first interpretable ultrasound acquired no sooner than 15 min after the start of hemodialysis. This 15 min delay ensured that the baseline SpHb measurements were not affected by the 250 cc of 0.9% sodium chloride solution that was used to prime the extracorporeal dialysis circuit. Subsequent ultrasounds were acquired every 15 to 30 min. This flexibility in timing was required to prevent interference with routine clinical care. The last ultrasound was acquired 2 h after the first; over this duration we anticipated an average drop in blood volume of 5%, given typical ultrafiltration rates (Santoro et al 1998, Krepel et al 2000).

The median SpHb from each 15 min interval was converted to RBV_SpHb by the following formula (Dasselaar et al 2005):

$$\begin{align} \newcommand{\e}{{\rm e}} \displaystyle {\rm RB}{{{\rm V}}_{{\rm SpHb}}}=~\left[ \frac{{\rm SpH}{{{\rm b}}_{{\rm baseline}}}}{{\rm Median}\,{\rm SpHb}}-1 \right]\times 100.\nonumber \end{align} \tag{ 1 }$$

This formula illustrates that RBV_SpHb is the inverse of the relative change in SpHb: as hemoconcentration occurs, SpHb increases while RBV_SpHb decreases.

We refer to patterns of rising and falling measurements of ${\rm IV}{{{\rm C}}_{{\rm US}}}$ and RBV_SpHb over time as profiles. Separate previous reports found that profiles of both measurement types had three distinct shapes that reflected different fluid states at the start of hemodialysis (Lopot et al 1996, Katzarski et al 1997). Those with the least fluid overload (the most 'dry') had decelerating negative profiles. The profiles were negative because the rate of fluid removal by ultrafiltration outpaced the rate of refill from the interstitium; and they were decelerating because initially steep negative slopes flattened as ultrafiltration was reduced to avoid hypotension (Mitra et al 2002). Those with intermediate fluid overload had flat profiles because refill from the interstitium matched fluid removal by ultrafiltration and blood volume remained unchanged (Sinha et al 2010). Those with the most fluid overload (the most 'wet') had decelerating positive profiles (Agarwal and Weir 2010). The profiles were positive because fluid shifted from the interstitium upon reclining faster than it was removed by ultrafiltration; and they were decelerating because initially steep positive slopes flattened as this gravitational fluid shift later reached equilibrium (Hinghofer-Szalkay and Moser 1986). We anticipated that our participants would segregate into these three profiles for each type of measurement (${\rm IV}{{{\rm C}}_{{\rm US}}}$ and RBV_SpHb), and we tested the hypothesis that within each participant, the shape of the measurement profile based on ${\rm IV}{{{\rm C}}_{{\rm US}}}$ would match the shape of the profile based on RBV_SpHb.

We based our sample size calculation on the kappa statistic under the assumption that decelerating positive profiles would be half as common as the flat and decelerating negative profiles (Tchernodrinski et al 2015). In order to detect an increase of 20% above chance agreement between ${\rm IV}{{{\rm C}}_{{\rm US}}}$ and RBV_SpHb profiles with a power of 80% and a two-sided type 1 error of 5%, we estimated a target sample size of 50 participants (Hong et al 2014).

We used group-based trajectory modeling to group participants by their profile for each measurement type (Nagin 2005). This statistical technique defines subpopulations by the profiles that emerge from serial measurements, rather than by preconceived notions of how those groupings are defined. It is most useful when different profiles are hypothesized a priori (as in this case based on fluid status), but grouping variables that define the relevant subpopulations are unknown. To formally test whether three distinct profiles emerged as the most parsimonious description of our data, and whether these profiles were curvilinear in shape (as suggested by previous reports (Mitra et al 2002)), we compared group-based trajectory models using the Bayesian information criterion (Raftery 1995). Average profiles for all participants and for groupings based on profile shapes were depicted as quadratic functions with random trend mixed-effects regression models (Hedeker and Gibbons 2006).

To compare characteristics across profile groupings, we used tests for trend that do not assume linear relationships. For parametric data, we used a joint test of unbalanced, non-orthogonal contrasts from adjacent-level means generated by one-way analysis of variance (Davis 2010), and for nonparametric data, we used an extension of the Wilcoxon rank-sum test developed by Cuzick (1985).

To describe the strength of the linear associations between change in ${\rm IV}{{{\rm C}}_{{\rm US}}}$ and change in RBV_SpHb, we used Pearson's correlation coefficient. Kappa statistics were unweighted with confidence intervals that were bias-corrected with a bootstrap method using 1000 replications (Reichenheim 2004). All analyses were conducted with Stata, version 15.1 (StataCorp, College Station, TX), including group-based trajectory modeling, which was carried out with the traj (Jones and Nagin 2013) command from within Stata.

Seventy-three participants were enrolled, but 23 (32%) were excluded, as detailed in figure 1. Major reasons for exclusion were that IVC ultrasounds could not be acquired or interpreted, or SpHb measurements could not be interpreted due to persistently poor signal quality. The characteristics of the remaining 50 participants who were included in the agreement assessment are listed in table 1. Participants were predominantly Caucasian (96%), overweight (mean BMI 27 m kg⁻²), and on renal replacement therapy for over a year (median 15 months). Among 31 (62%) participants with available echocardiographic data, moderate or severe tricuspid regurgitation was present in a third, and the left-ventricle ejection fraction was preserved on average (mean 58%).

Zoom In Zoom Out Reset image size

Overall there was a total of 242 paired ${\rm IV}{{{\rm C}}_{{\rm US}}}$ and RBV_SpHb measurements; within each participant there was a mean of five paired measurements (range 2–7 measurements). As depicted in figure 2, individual trends in ${\rm IV}{{{\rm C}}_{{\rm US}}}$ and RBV_SpHb were highly variable so that the average trends among all participants were relatively flat. After a mean of 1.3 liters was removed by ultrafiltration, the average ${\rm IV}{{{\rm C}}_{{\rm US}}}$ decreased by only  −1.0 mm (95% CI  −1.9 to  −0.2 mm) while the decrease in RBV_SpHb was not statistically different than zero (−1.1%, 95% CI  −2.7 to  +0.5%). There was no linear association between IVC_US change or RBV_SpHb change, either for the entire 2 h (r  =  −0.04, 95% CI  −0.32 to  +0.24) or when each hour was considered separately (figure 3).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Group-based trajectory modeling categorized IVC_US and RBV_SpHb profiles into three curvilinear shapes (figure 4). However, the agreement between profile shapes of each measurement type within participant was no better than that expected by chance (kappa  −0.1, 95% CI  −0.3 to  +0.1; table 2). For example, 85% (11 of the 13 participants) who were categorized as having a decelerating negative IVC_US profile were also categorized as not having a decelerating negative RBV_SpHb profile (eight were categorized as having a flat RBV_SpHb profile and three as having a decelerating positive RBV_SpHb profile; appendix, figure A1). There were minimal differences in baseline characteristics across participants when grouped either by IVC_US or RBV_SpHb profiles (appendix, table A1, and appendix, table A2).

Zoom In Zoom Out Reset image size

Table 3 further illustrates how IVC_US and RBV_SpHb provided different information. No trend in maximal changes of RBV_SpHb measurements was observed across participants grouped by IVC_US profiles (p -value 0.80). Nor was there a trend in maximal changes of IVC measurements across participants grouped by RBV_SpHb profiles (p -value 0.26). Of the two measurement types, only RBV_SpHb profiles were associated with the rate of ultrafiltration, which is the most proximate cause of blood volume change during hemodialysis: when grouped by RBV_SpHb profiles, participants' net ultrafiltration rates were higher with more negative profiles (p -value for trend 0.02).