BACKGROUND. Assessment of dehydration in young children currently depends on clinical judgment, which is relatively inaccurate. By using digital videography, we developed a way to assess capillary-refill time more objectively.
OBJECTIVE. Our goal was to determine whether digitally measured capillary-refill time assesses the presence of significant dehydration (≥5%) in young children with gastroenteritis more accurately than conventional capillary refill and overall clinical assessment.
METHODS. We prospectively enrolled children with gastroenteritis,1 month to 5 years of age, who were evaluated in a tertiary-care pediatric emergency department and judged by a triage nurse to be at least mildly dehydrated. Before any treatment, we measured the weight and digitally measured capillary-refill time of these children. Pediatric emergency physicians determined capillary-refill time by using conventional methods and degree of dehydration by overall clinical assessment by using a 7-point Likert scale. Postillness weight gain was used to estimate fluid deficit; beginning 48 hours after assessment, children were reweighed every 24 hours until 2 sequential weights differed by no more than 2%. We compared the accuracy of digitally measured capillary-refill time with conventional capillary refill and overall clinical assessment by determining sensitivities, specificities, likelihood ratios, and area under the receiver operator characteristic curves.
RESULTS. A total of 83 patients were enrolled and had complete follow-up; 13 of these patients had significant dehydration (≥5% of body weight). The area under the receiver operator characteristic curves for digitally measured capillary-refill time and overall clinical assessment relative to fluid deficit (<5% vs ≥5%) were 0.99 and 0.88, respectively. Positive likelihood ratios were 11.7 for digitally measured capillary-refill time, 4.5 for conventional capillary refill, and 4.1 for overall clinical assessment.
CONCLUSIONS. Results of this prospective cohort study suggest that digitally measured capillary-refill time more accurately predicts significant dehydration (≥5%) in young children with gastroenteritis than overall clinical assessment.
- preschool child
- diagnostic techniques and procedures
- sensitivity and specificity
Assessment for the presence and severity of dehydration at the bedside is still dependent on findings derived from clinical examination.1–3 The World Health Organization (WHO), the American Academy of Pediatrics, and a recently published systematic review concur that a combination of signs, although relatively inaccurate, are the best tools available to the practicing clinician.1–3 Reliance on clinical judgment, however, can result rarely in significant morbidity and mortality and, more commonly, in excessive utilization of health care resources.4–7
The best single sign for predicting significant dehydration ≥5% in children is prolonged capillary refill.3 However, this sign, although relatively specific (0.85; 95% confidence interval [CI]: 0.72 to 0.98), is not sensitive (0.60; 95% CI: 0.29 to 0.91).3 We hypothesized that an instrument that accurately measures capillary refill would improve assessment of dehydration.
By using digital videography, we developed a novel, noninvasive way to assess capillary-refill time, referred to as digitally measured capillary-refill time (DCRT). The objective of this study was to determine whether this measure of capillary bed refilling time predicts significant dehydration (≥5% of body weight) in young children with acute gastroenteritis more accurately than the assessment by experienced clinicians who were using conventional capillary refill and their overall clinical assessment.
This prospective cohort study was conducted at a university-affiliated, urban pediatric emergency department, the Alberta Children's Hospital in Calgary, Canada. We enrolled 2 types of patients into separate phases. In the first phase, we assessed DCRT in children who were not dehydrated to establish the reference range for this test. In the second phase, we enrolled children with acute gastroenteritis who were assessed by clinicians to have at least mild dehydration. The study was approved by our institutional human ethics board, and parents of eligible patients provided written informed consent.
In the first phase, we enrolled children 1 month to 5 years of age if they suffered a minor injury, such as a superficial laceration or minor head injury, and were otherwise healthy without a recent history of vomiting, diarrhea, or fever; we excluded children with a history of chronic disease or failure to thrive. Children were enrolled into this phase at the convenience of the study investigator (Dr Shavit).
In the second phase, we enrolled children between 1 month and 5 years of age with a history of diarrhea (with or without vomiting) for ≤5 days and who were judged by the emergency department triage nurse to have some degree of dehydration. We excluded children with a history of cardiovascular or renal disease, as well as those who were judged by the triage nurse to require emergent medical intervention. The study investigator was notified by the triage nurse about children potentially eligible to be enrolled into this phase.
For both study phases, the study investigator determined whether children met enrollment criteria, obtained consent, and conducted all study measurements, including the index and reference tests. Immediately after enrollment and before any treatment, study patients were weighed dressed only in a dry diaper or underwear on a calibrated portable electronic scale (Infant Checker; Madela Inc, McHenry, IL). The ambient air temperature of the examination room was measured by using a mercury thermometer (Electro-Therm; Newco, Inc, Florence, SC); the child's core temperature was measured by using an infrared tympanic thermometer (First Temp Genius, Model 3000A; Sherwood Medical, St Louis, MO); and the child's fingertip skin temperature was measured by using an infrared skin surface thermometer (Dermatemp DT1001; Exergen, Watertown, MA).
For both study phases, the DCRT was determined as follows. Children were placed in a supine position, and one of their hands was raised slightly above the level of the heart. The study investigator used a small digital video camera (Zoltrix Eagle Cam USB CMOS camera; Kowloon Bay, Hong Kong) with customized graphic software to film one of the child's fingertips. The camera was fixed by a mechanical holder between 3 and 5 centimeters from this fingertip. The camera's digital video file was automatically converted by the customized software to a “frame-by-frame” analysis. This allowed the fingertip's exact color characteristics to be determined. Subsequently, the investigator lightly pressed a small smooth-tipped rod (with a contact surface of 0.5 mm2) against the end of this fingertip for 5 seconds; he then abruptly released it. After release of the rod, the software compared each subsequent frame with the initial “precompression” frame until an exact color match was achieved. The first frame that matched the precompression frame was termed the “recovery frame.” The software then automatically calculated the time between rod release and the recovery frame. This time, the DCRT, was displayed on a small digital screen hooked up to the camera.
To establish the reproducibility of the index test (or the ability of the test to be accurately replicated), the DCRT was repeated at least twice on 1 fingertip and again on a different finger, all within 30 to 60 seconds of each other. The first measurement was used to represent the child's DCRT.
After the initial assessment in the second phase only, the child's staff pediatric emergency physician was asked to estimate their degree of dehydration by using a 7-point Likert scale (very mild to very severe)8 and to determine their capillary-refill time by using standard clinical techniques (<2 seconds or ≥2 seconds).9–12 All staff physicians (21 total) were trained and board certified in pediatrics or emergency medicine and worked as staff between 1 and 18 years. Once all measurements were completed, each patient received therapy as judged necessary by the staff physician. Staff physicians did not have access to the DCRT measurements.
The reference (or gold) standard for determining a child's percentage fluid deficit is to subtract a child's weight at initial assessment from a stable postillness weight.3,13 To determine this latter weight, patients enrolled in the second phase were weighed using the same scale 48 and 72 hours after enrollment, and, as necessary, every 24 hours until both the child's diarrhea and vomiting had resolved and a stable weight was achieved. The primary investigator assured follow-up by making home visits to assess and weigh all enrolled patients. A stable postillness weight was taken when 2 consecutive weights differed by less than 2%. The mean of these 2 weights was used to determine the postillness weight gain. Each child's fluid deficit was categorized on the basis of the WHO's definition as either no dehydration (<5%) or significant dehydration (≥5%).2
For the study's first phase, we planned to enroll at least 50 patients to adequately assess the impact of variables such as age, ambient temperature, skin temperature, and core temperature because multivariate analysis requires at least 10 patients for each independent variable.
In the study's second phase, we planned to enroll at least 65 patients to allow us to detect an absolute difference in the sensitivity of the DCRT versus conventional capillary refill of 25% for detecting significant dehydration ≥5%. This assumed an α of .05, a β of 0.20, and the sensitivity of conventional capillary refill to be 50%.
To assess test-retest reliability (ie, the reproducibility) of DCRT, we plotted points representing repeat measurements, with X as the initial measurement and Y as the subsequent. We then calculated the intraclass correlation coefficient; this is similar in interpretation to the ordinary (Pearson) correlation coefficient but measures correlation about the line of equality (ie, Y = X). Values of the intraclass correlation coefficient >0.8 are usually considered to represent very good reproducibility.
To assess whether patient age and core, skin, and ambient temperatures confounded the diagnostic accuracy of DCRT, we plotted values and calculated linear regressions.
To assess the clinical performance of the 3 measurements (DCRT, conventional capillary-refill time, and overall clinical assessment), we examined them in relation to percent dehydration (<5% vs ≥5%), which we took as the gold standard. Before we could apply conventional measures of diagnostic accuracy (eg, sensitivity, specificity) we first examined receiver-operator characteristic (ROC) curves to estimate reasonable cut points for clinical decision-making. Reviewing briefly, if we consider values at or above a cutoff, C, as test-positives (on the assumption that high values of T are indicative of disease), then the sensitivity of the associated diagnostic rule is the proportion of true-positives, the test value of which, T, is C or higher. Similarly, the observed specificity of the test is the proportion of true-negatives with T < C. The ROC curve is a plot of the values of sensitivity (on the y-axis) versus values of one specificity (on the x-axis) across all possible choices of C (2 such curves are plotted in Fig 1)
Because low values of C will produce low sensitivities and high specificities and vice versa for high values of C, the ROC curve of a test will generally resemble the upper left-hand arc of a circle, rising from a value of 0 to a maximum of 1 on the y-axis as one moves from 0 to 1 on the x-axis. The closer the curve comes to the upper left-hand corner of the plot, the better the test performance. A test whose ROC curve is a straight line from lower left-hand to upper right-hand is basically useless. To capture this quantitatively, one can calculate the area under the ROC curve (AUC) . Values of the AUC close to 1 are optimal. A value of 0.5 corresponds to a useless test, whereas values <0.5 denote a worse-than-useless test (eg, a test is wrong 100% of the time). A pragmatic choice of C, if sensitivity and specificity are judged to be of equal importance, is to take the value of C yielding the point closest the upper left corner.
Having chosen cutoffs for each of the 3 tests, we show the associated sensitivity and specificities, and positive and negative likelihood ratios (LR+ and LR−, respectively) with 95% CIs. The latter are arguably the most relevant to clinical decision-making, because if a patient has a probability, P, of being a true-positive on the basis of factors observed before testing, LR+ and LR− can be combined with P to provide posttest probabilities, as follows. The posttest odds for positivity for a person can be calculated by multiplying the pretest odds, P/(1 − P), by the LR+ or LR− (according to whether their test result is positive or negative). The posttest probability can then be calculated as (posttest odds)/(1 + posttest odds). We have also assessed the statistical significance of apparent differences in test performance using McNemar's test for paired binary data.
In November and December 2000, we enrolled 65 children into the study's first phase. Between January 2001 and April 2002, we enrolled 87 children into the study's second phase. The children enrolled were at the convenience of the study investigator. In addition, the total number of patients evaluated in our institution with simple injuries and gastroenteritis during these time periods was not tracked. The median ages of the first and second phases were 30 months (interquartile range [IQR]: 17–47 months) and 18 months (IQR: 11–34 months), respectively. Of the 87 enrolled into the second phase, 83 (95%) had complete follow-up. One family withdrew consent before final assessment, and 3 families were lost to follow-up; none of these patients' results are shown (Fig 2).
All staff physicians' overall clinical assessments for patients in the second phase were completed within 5 to15 minutes of DCRT measurements and before treatment with either oral or intravenous fluids. The range of follow-up required for the 83 children to resolve gastrointestinal symptoms and achieve a stable weight was 2 to 5 days. Of the 83 patients enrolled into the second phase who had complete follow-up, 70 (84%) had a <5% fluid deficit, and 13 (16%) met the WHO definition of dehydration (≥5%); 12 children had fluid deficits ranging from 5% to 8%, and 1 child had an 11% fluid deficit. Only 1 of 83 children was admitted to the hospital; the rest were discharged from the hospital.
For patients in the first phase, the DCRT ranged between 0.2 and 0.4 seconds (median: 0.2 seconds; IQR 0.2–0.3 seconds), and for patients in the second phase, DRCT ranged between 0.2 and 0.6 seconds (median: 0.3 seconds; IQR: 0.2–0.35 seconds). For patients in the second phase, a strong relationship between a child's fluid deficit and their DCRT was evident (Pearson correlation coefficient: 0.75; P < .001). Test reproducibility for the DCRT on 2 different fingertips performed within 60 seconds of each other, assessed by intraclass correlation, was 1.00 for the first phase and 0.99 for the second phase.
Phase 2 Estimates
The AUC relative to the presence of dehydration ≥5% for DCRT and overall clinical assessment was 0.99 (95% CI: 0.97 to 1.00) and 0.88 (95% CI: 0.78 to 0.96), respectively (comparison of the 2 test methods, P = .013; see Fig 1). We could not generate an ROC curve for conventional capillary-refill time because only 1 cutoff point was determined (less than or greater than 2 seconds).
The optimal cutoff value demonstrated by the AUC for DCRT was ≥0.4 seconds and for overall clinical assessment was ≥4 seconds. The sensitivity for predicting dehydration (>5%) was 100% (95% CI: 75% to 100%) for DCRT; 77% (95% CI: 46% to 95%) for overall clinical assessment; and 54% (95% CI: 25% to 81%) for conventional capillary-refill time. The percentage difference in sensitivity between DCRT and conventional capillary refill was 46% (95% CI: 11% to 81%; P = .03); between DCRT and overall clinical assessment was 23% (95% CI: −8% to 54%; P = .25); and between conventional capillary refill and overall clinical assessment was −23% (95% CI: −69% to 23%; P = .33).
The specificity for predicting dehydration ≥5% was 91% (95% CI: 82% to 97%) for DCRT; 88% (95% CI: 75% to 93%) for conventional capillary refill; and 81% (95% CI: 70% to 90%) for overall clinical assessment. The percentage difference n specificity for DCRT and conventional capillary refill were 6% (95% CI: −4% to 15%; P = .22); for DCRT and overall clinical assessment were 10% (95% CI: −2% to 22%; P = .08); and for conventional capillary refill and overall clinical assessment were 4% (95% CI: −8% to 17%; P = .31).
Positive likelihood ratios were 11.7 (95% CI: 5.4 to 22) for DCRT, 4.5 (95% CI: 2.0 to 10.4) for conventional capillary refill, and 4.1 (95:% CI 2.3 to 7.4) for overall clinical assessment. Negative likelihood ratios were 0 (estimated upper 95% confidence limit: 0.6) for DCRT, 0.52 (95% CI: 0.29 to 0.95) for conventional capillary refill, and 0.28 (95% CI: 0.10 to 0.77) for overall clinical assessment. A scatterplot illustrating DCRT relative to percent dehydration is shown in Fig 3.
Potential Confounding Variables
The ambient room temperatures ranged between 20.3 and 25.6°C (mean: 22.9 ± 0.7); patients' fingertip skin temperatures ranged between 23.1 and 34.7°C (mean: 28.9 ± 2.5); and tympanic temperatures ranged between 35.2 and 39.9°C (mean: 37.3 ± 1.0). A graphic exploration of the relationship between DCRT, age, and the 3 measured temperatures for phase 2 patients showed no-to-modest associations. The Pearson correlation coefficient between DCRT and age was 0.23 (P = .04); between DCRT and ambient temperature, it was −0.18 (P = .10); between DCRT and core temperature, it was 0.03 (P = .81); and between DCRT and skin temperature, it was –0.34 (P = .001). In multiple regression analysis, only DCRT and overall clinical assessment were significantly associated with fluid deficit.
We found in our small cohort of young children with gastroenteritis, that the DCRT was more accurate, as assessed by the AUC, at determining the presence of significant dehydration (≥ 5%) than the overall clinical assessment of experienced pediatric emergency physicians. DCRT was more sensitive than conventional capillary refill but not more specific. However, the 95% CIs for the sensitivity, specificity, and positive likelihood ratio of DCRT relative to overall clinical assessment were overlapping, at least in part because our cohort included only 13 children with dehydration.
We chose the WHO's cutoff of 5% as distinguishing between significant and nonsignificant dehydration. because we thought it was practical and clinically meaningful.2 Although transformation of a continuous measure into a simple dichotomy results in some loss of information, it allowed us to calculate sensitivities, specificities, and likelihood ratios, which are more readily understood than continuous measures of test accuracy.
Because we wanted to compare DCRT with standard clinical assessment of experienced pediatric physicians, we chose physicians' clinical assessment that used a Likert scale as the comparator because it is our experience that physicians do not formally score and tally up a specific set of clinical signs to arrive at their assessments. It is important to note, however, that our use of an overall clinical assessment does not allow us to compare DCRT with individual clinical signs, which is the usual method for predicting dehydration in studies.3,13,14 However, we found the accuracy of physicians' overall clinical assessment (as assessed by the AUC: 0.88) to be comparable to the accuracy of a compilation of 4 and 10 individual signs shown by Gorelick13 (AUCs of 0.90 and 0.91, respectively).
We found a substantially different cutoff time for DCRT (0.4 seconds) than the routine cutoff used for capillary refill (2.0 seconds).9 Conventional capillary refill is assessed by a variety of methods, such as compression of the finger pulp, the nailbed, or the chest wall, and usually involves firm compression of the examined body part by the examiner's thumb or index finger for up to 5 seconds.9–12 In contrast, we lightly compressed the finger tip with a small, pointed rod, which emptied a much smaller part of the capillary bed. We think this is why the refill time for DCRT is so much shorter than conventional capillary refill.
Our results suggest that DCRT is more sensitive for assessing significant dehydration (≥5%) than conventional capillary refill without any apparent loss of specificity. We think the reason for this may be that the technique minimizes subjective judgment in the assessment of capillary refill, which, in turn, results in substantially better precision.
Unlike previous studies that have suggested that ambient, skin, or core temperatures can, to some degree, affect conventional assessment of capillary refill,9,10,12 we found no significant correlation between any of these temperatures and DCRT. It is unclear to us why we found comparatively modest correlations between DCRT and ambient, skin, and tympanic temperatures. It may result, at least in part, from DCRT's greater precision than conventional capillary refill. However, given the limited size of our cohort, additional examination of the relationship between temperatures and this new technique is warranted.
There are several limitations to our study, some of which have already been pointed out. First, our study sample is small and includes a relatively small percentage of significantly dehydrated children. Second, we only enrolled patients if they were assessed by a triage nurse to have some degree of dehydration. In addition, we made no effort to track potentially eligible patients, thus we cannot report how many were missed and whether those enrolled were comparable to those not enrolled. Our focus on such a selected population may have caused spectrum bias. Third, we conducted our study at only 1 tertiary hospital, which potentially threatens the external validity of our results. Consequently, for all these reasons, our results must be interpreted with caution.
Furthermore, this novel device's current design makes it impractical for routine use in clinical settings. Even if a more advanced prototype of this device does decrease the number of children who are inaccurately diagnosed, additional study will be necessary to determine whether such a device would actually decrease unnecessary health care utilization or improve health outcomes.
Nonetheless, our tentative findings, that digitally aided measurement of capillary-refill time seems to be more accurate for diagnosing significant dehydration (≥5%) in young children than routine overall clinical assessment, are sufficiently strong to warrant additional technical development and study of this concept.
The electronic scale was purchased with general research funds provided by Dr Nijssen-Jordan. The infrared skin surface thermometer was provided by the Department of Anesthesiology at Alberta Children's Hospital, Canada.
- Accepted August 30, 2006.
- Address correspondence to David W. Johnson, MD, Room C4 643, Alberta Children's Hospital, 2888 Shaganappi Trail NW, Calgary, Alberta, Canada T3B 6A8. E-mail:
Dr Shavit conceived of the idea for the development of the DCRT technique and the original study question; he enrolled all patients and gathered, entered, and verified all data; he completed the initial draft of the manuscript; and he bears overall responsibility for study content. Dr Brant established the procedures for statistical analysis, analyzed the data, and critically reviewed the manuscript. Dr Nijssen-Jordan critically reviewed the study protocol and manuscript. Dr Galbraith critically reviewed the study design and manuscript. Dr Johnson designed the study and was the principal author of the study protocol and manuscript, and he shares overall respondibility for study content.
Financial Disclosure: Dr Shavit, the principal investigator, conducted this study during his fellowship in Pediatric Emergency Medicine at the University of Calgary. Dr Shavit conceived of the idea for the study, the DCRT technique, and developed the customized algorithm used to analyze the DCRT. After completion of this research project, Dr Shavit presented his idea for the DCRT to the faculty at Technion (the Israel Institute of Technology), and an incubator company was formed. Dr Shavit no longer has any financial interests in this company.
This work was presented at the annual meeting of the American Academy of Pediatrics; October 18, 2002; Boston, MA.
Dr Brant's current address is Department of Statistics, University of British Columbia, Vancouver, British Columbia, Canada.
Dr Shavit's current address is Meyer Children's Hospital, Rambam Medical Center, Technion University, Haifa, Israel.
- ↵American Academy of Pediatrics, Provisional Committee on Quality Improvement, Subcommittee on Acute Gastroenteritis. Practice parameter: the management of acute gastroenteritis in young children. Pediatrics.1996;97 :424– 426
- ↵World Health Organization. The Treatment of Diarrhea: A Manual for Physicians and Other Senior Health Workers. Geneva, Switzerland: World Health Organization; 1995
- Santosham M, Keenan EM, Tulloch J, Broun D, Glass R. Oral rehydration therapy for diarrhea: an example of reverse transfer of technology. Pediatrics.1997;100(5) . Available at: www.pediatrics.org/cgi/content/full/100/5/e10
- ↵Bellemare S, Hartling L, Wiebe N, et al. Oral rehydration versus intravenous therapy for treating dehydration due to gastroenteritis in children: a meta-analysis of randomised controlled trials. BMC Med.2004;15 :11
- ↵Streiner DL, Norman GR. Health Measurement Scales: A Practical Guide to Their Development and Use. Oxford, United Kingdom: Oxford University Press; 1995
- ↵Gorelick MH, Show KN, Baker D. Effect of ambient temperature on capillary refill in healthy children. Pediatrics.1993;92 :699– 702
- ↵Gorelick MH, Shaw KN, Murphy KO. Validity and reliability of clinical signs in the diagnosis of dehydration in children. Pediatrics.1997;99(5) . Available at: www.pediatrics.org/cgi/content/full/99/5/e6
- Copyright © 2006 by the American Academy of Pediatrics