Original Contribution

A better way to estimate adult patients’ weights^?

Brian W. Lin MD^a,?, Douglas Yoshida MD^b, James Quinn MD^b, Matthew Strehlow MD^b

^aStanford/Kaiser Emergency Medicine Residency, San Francisco, CA 94116, USA

^bDivision of Emergency Medicine, Department of Surgery, Stanford University Hospital, Stanford, CA 94305, USA

Received 17 June 2008; revised 3 August 2008; accepted 3 August 2008

Abstract

Objective: In the emergency department (ED), adult patients’ weights are often crudely estimated before lifesaving interventions. In this study, we evaluate the reliability and accuracy of a method to rapidly calculate patients’ weight using readily obtainable anthropometric measurements. We compare this method to Visual estimates, patient self-report, and measured weight.

Methods: A convenience sample of adult ED patients in an academic medical center were prospectively enrolled. Midarm circumference and knee height were measured. These values were input in to equations to calculate patients’ weights. A physician and nurse were then independently asked to estimate the patients’ weights. Each patient was asked to report his/her own weight before being weighed. Calculated weights using the above equations, visual estimates, and patient reports were compared with actual weights by determining the percentage accurate within 10%. The intraclass correlation coefficient was used to determine the reliability of the estimates with respect to actual weights.

Results: Weight was determined within 10% accuracy of actual weight in 69% (95% confidence interval, 63-75) of calculated estimates, 54% (48-61) of physician estimates, 51% (44-57) of nurse estimates, and 86% (81-90) of patient estimates. The weight estimation tool calculated weights more accurately in males (74%, 65-82) than females (65%, 56-73). An analysis of errors revealed that when estimates were inaccurate, approximately half were overestimates and half were underestimates. The correlation coefficient between the calculated estimates and actual weights was 0.89. The correlation coefficient of actual weights with respect to physician estimates, nurse estimates, and doctor’s estimates were 0.85, 0.78, and 0.95, respectively.

Conclusions: This technique using readily obtainable measurements estimates weight more accurately than ED providers. The technique correlates well with actual patient weights. When available, patient estimates of their own weight are most accurate.

(C) 2009

Introduction

^? Work previously presented at Mediterranean Emergency Medicine Congress, Sorrento, Italy, September 2007.

* Corresponding author.

E-mail address: [email protected] (B.W. Lin).

It is estimated that Medication errors cost billions of dollars and harm 1.5 million people per year [1]. Although the exact incidence is difficult to determine, it is estimated based on previous studies that per 1000 orders, 0.61 to 53 prescribing errors are made [1]. These include errors in weight-based dosing of medications. Weight-based medication therapy is

0735-6757/$ – see front matter (C) 2009 doi:10.1016/j.ajem.2008.08.018

integral in caring for patients in critical situations. Many lifesaving treatments–including thrombolytics, anticoagu- lants, vasopressors, and intravenous fluids–are dosed based on a patient’s weight. The safety and efficacy of these interventions is adversely affected by inappropriate dosing, which can easily occur if the wrong weight is estimated and used.

Limitations of resources, physical space, and time often preclude providers from obtaining measured weights. Although patients can usually accurately predict their weight, a large proportion of patients who are critically ill are unable to provide this important information. Further- more, health care providers have been shown to poorly estimate weights [2-4]. Corbo et al [4] found that physicians and nurses were able to estimate a patient’s weight within 10% of their actual weight only approximately 50% of the time. A more recent study by Kahn et al [5] showed that providers were able to estimate within 5% of true weight in only 33% of estimates. Physician accuracy did not improve with experience or specialty [4].

A rapid, reliable weight estimation tool could be invaluable to reduce errors in the delivery of various medications and fluids used in the emergency department (ED). Anthropometric parameters such as midarm circum- ference, knee height, demi-span, and other body measure- ments have been used to create nomograms and equations derived from statistical modeling to estimate patients’ weights. Generally, these tools have been developed for use in nonemergent settings. Furthermore, no such tool has gained wide acceptance nor has any been tested specifically and prospectively validated in an ED population. In this study, we selected a practical, relatively simple tool and tested it in an ED population.

Materials and methods

Study subjects

A prospective cohort of consecutive, eligible, consented patients at a tertiary care academic facility were enrolled for an 11-month period when the study principal investigator was present and available to enroll patients in the ED. We included ambulatory and nonambulatory adult patients who agreed to be weighed in the ED or upon admission. This included obtunded, lethargic, and Unconscious patients, who were consentable by health care proxy. We excluded nonambulatory patients who were discharged from the ED who could not be easily weighed in a timely fashion. The institutional review board at Stanford University (Stanford, Calif) approved the study protocol.

Interventions

Upon enrollment, the physician and nurse caring for the patient were asked to estimate the patient’s weight. The

patient was also asked to report his or her own weight. Anthropometric measurements were then obtained using a flexible tape measure. We selected a weight estimation tool authored by Ross Laboratories, using measurements of knee height and midarm circumference, for evaluation. To reproduce the method created by Ross Laboratories most accurately, we reviewed the instructions from the Ross Laboratories instruction manual in defining anthropometric parameters [6]. Measurements of midarm circumference were obtained by measuring the circumference of the arm at the midpoint of the upper extremity. This was determined by first palpating the humeral head as a proximal landmark. The lateral condyle of the distal humerus was identified as a distal landmark. The midpoint between these landmarks was located and used to define the midpoint of the arm. The tape measure was wrapped around the midpoint of the arm such that it contacted the skin at all points but did not compress the soft tissue. Measurements of knee height were defined proximally by the thigh prominence with the knee bent at 90? and distally by the plantar aspect of the foot at the heel with the ankle bent at 90?.

Actual weight was then obtained using 1 of 3 scales, a portable digital scale (used at the bedside for patients weighing b200 lb), a Health O’Meter scale (used for patients weighing (approximately 200-300 lbs), and a Scale-Tronix. The use of the Health O’Meter scale was a balance-based scale, used to ensure accuracy in heavier patients, as the portable scale was not able to accurately measure weight beyond 200 pounds. The Scale-Tronix was a digital scale that included arm rail-assists that were calibrated into the final reading. This scale was used for patients who were minimally ambulatory and required some support while standing. The 3 scales were calibrated to each other on every day of the study. Patients were weighed wearing cloth patient gowns and undergarments; some patients wore other articles of clothing that they refused to remove, but all patients were barefoot and were asked to remove accessories such as jewelry, keys, wallets, and cell phones. Patients who could not be weighed in the ED because they were unable to stand were weighed using electronic bed scales upon admission. These weights were obtained by inpatient nursing staff and were not calibrated to our scales. Weight measurements obtained using the portable digital scale were considered the preferred standard for the study, but measurements from inpatient admission were accepted as a surrogate.

Patients, physicians, and nurses were blinded to the actual weight and to estimates by other individuals. Nurses and physicians were not given feedback on their estimations during the study period to prevent bias caused by feedback learning. Estimates in pounds or kilograms were permitted to allow physicians and nurses to provide their best possible estimate. All estimates were converted to kilogram weights for data analysis and rounded to 3 significant digits.

Other prospective data on our patient population was also collected, including age, ethnicity as reported by the patient, and any significant medical history (organ transplant, major

surgery, congestive heart failure, liver disease) as reported by the patient.

We used the following equations produced by Ross Laboratories to calculate weight:

male^‘s weight(kg)= knee height x 1.10 + MAC(cm)

x 3.07-75.81

and

female^‘s weight(kg)= knee height x 1.01 + MAC(cm)

x 2.81-66.04

estimation error was defined as follows:

[absolute value of (estimated weight – actual weight)/ actual weight].

Estimates were then categorized as within 10% or greater than 10% error. We chose this cutoff because 10% is the acceptable margin of error commonly cited in the literature for thrombolytics and anticoagulants.

Outcome measures and statistical methods

Before the study, we determined that a 10% improve- ment in accuracy at estimating weight within 10% of actual weight would be a clinically important difference. Assu- ming a visual estimation accuracy of 50% based on previous studies, we calculated a sample size of 325 would be required to detect a 10% difference in accuracy using an ? = .05 and a ? = .02.

Percent accuracy and 95% confidence intervals were calculated using JavaStat Software. To assess the strength of the relationship between actual weights and estimations, we calculated the intraclass correlation coefficient using SPSS software (SPSS, Chicago, Ill). The intraclass correlation coefficient is a measure of reliability. Based on regression analysis, it measures the extent to which the relationship between 2 variables (eg, calculated weight and actual weight) can be described by a straight line [7]. When a perfect fit is achieved, the correlation coefficient equals

1.0. A correlation coefficient of 0.85 has been cited in the

Table 1 Patient demographics and clinical characteristics

Male		Female	Total
No. of patients	111	123	235
Age (y)	51.5	47.8
Age range (y)	18-97	18-91
Average weight (kg)	80.3	67.7
SD (kg)	22.5	16.5
Race
White (%)			55
Hispanic (%)			21
Asian (%)			11
African American (%)			3
Other (%)			7

Fig. 1 Percentage of estimates within 10% of actual measured weight. Calculated weights using the weight estimation tool are more accurate than physician and nurse estimates, whereas patient self-estimates are the most accurate.

literature as being acceptable for tests used to make Clinical decisions about individuals [8].

Results

Over the course of the study, 274 patients were asked to participate. Of these, 39 refused or could not be consented and 235 patients were enrolled. The demographic and clinical characteristics are summarized in Table 1.

Fig. 1 summarizes the key results of our study. This weight estimation tool predicted weight within 10% accuracy of a patient’s actual weight in 69% (95% confidence interval [CI], 63%-75%) of patients, which was significantly better than estimates by physicians and nursing staff, at 54% (95% CI, 48%-61%) and 51% (95% CI, 44%-57%),

respectively. Overall estimates by patients were the most accurate, with patients estimating their own weight within 10% accuracy 86% (95% CI, 81%-90%) of the time.

We used different formulas to calculate weight for men and women, as given by the Ross Laboratories method. Interestingly, examining the subgroups of men and women independently, we noted the calculation performed better in male patients. In men, the calculation predicted weight within 10% accuracy of a patient’s actual weight in 74%

Fig. 2 Subjects whose weight estimates were greater than 10% inaccurate using the weight estimation tool. Classified by degree of inaccuracy and whether weight was overestimated or underestimated.

Fig. 3 Graphical representation of the intraclass correlation coefficient, demonstrating reliability of calculated weights using weight estimation tool compared with actual measured weights.

(95% CI, 65-82) of patients. In women, the calculation predicted weight within 10% accuracy of a patient’s actual weight in 65% (95% CI, 56-73) of patients.

We also performed an analysis of errors (Fig. 2). We sought to examine if, in cases where the tool did not accurately estimate weight, most were underestimates or overestimates. We found that when the margin of error was 16% to 20% of actual weight, 19 cases were overestimates and 17 cases were underestimates. When the margin of error was 21% to 30%, 13 cases were overestimates and 9 were underestimates. Only 4 subjects had a margin of error greater than 30%. Among these patients, 3 were overestimates and 1 was an underestimate. There was no statistically significant difference in the number of overestimates compared with underestimates, regardless of the margin of error.

There was strong agreement between the weight estimation tool and actual weight. The intraclass correla- tion coefficient between calculated weights and actual weights was 0.89. This is graphically illustrated in Fig. 3. The intraclass Correlation coefficients between actual weight and other estimates obtained were as follows: patient self-estimates, 0.95; physician estimates, 0.85;

nurse estimates, 0.78.

Discussion

Our study showed that self-report is the most accurate method of weight estimation, and if that is not possible, a simple weight estimation tool can significantly improve the accuracy of physician and nurse estimates by 15%. This improved accuracy has potentially profound implications for patient safety by avoiding adverse and potentially lethal medication errors.

Of provider estimates, our study shows provider accuracy that is similarly poor as compared with other studies. In the study of Corbo [4], 50.4% of physician estimates and 49.6% of provider estimates were within 10% of true weight. Although not significant, our providers trended toward slightly more accurate estimates. Menon and Kelly [9] found that when held to a higher standard, nurses were only 44% accurate and physicians were only 33% accurate at estimating weight within 5%.

These results seem more worrisome as applied to a common clinical scenario encountered in one of our associated hospitals. Imagine a confused, 75-kg man unable to provide his weight whose electrocardiogram reveals an ST-segment elevation myocardial infarction. The patient cannot be accurately weighed in a timely fashion. The hospital lacks PCI capabilities. This patient requires a thrombolytic such as tenectaplase. The proper dose of this agent for this patient is 40 mg, where incremental changes in dose are made for patients weighing less than 70 kg or more than 80 kg. Based on our data, there is an approximately 50% chance this patient will be underdosed or overdosed if his dosing is based on a nurse or physician estimate. This includes risks of complications from overdosing, such as the risks of bleeding, especially Intracranial bleeding and hemorrhagic stroke. It also includes problems related to underdosing such as failure to achieve efficacious thrombo- lysis. With the weight estimation tool used in this study, the absolute risk reduction of overdosing or underdosing would be 15%, for a number needed to treat equal to 6.7. Thus, for every 7 patients where this method is used rather than a provider visual estimate, one case of incorrect dosing of potentially harmful medications would be prevented.

We also noted that the weight estimation tool was more accurate in estimating the weights of male patients than female patients. Different formulas provided by the Ross Laboratories method for men and women account for this difference. It may be that the formula provided for female patients has poorer accuracy and precision than that provided for male patients, and a better formula needs to be derived. We performed an analysis of errors to determine if, when inaccurate, the method was more likely to overestimate or underestimate patients’ weights. We also sought to determine what the margin of error was in these cases. Our method trended toward overestimating weight slightly more often than underestimating weight. However, the differences did

not reach statistical significance.

Our study has some limitations that warrant discussion. First, we acknowledge that although Ross Laboratories actually provides multiple formulas for use in several patient subsets, we selected only one for males and one for females for clinical evaluation. We did this to more realistically approximate an ED setting. In their equation set, race is dichotomized into “white” and “black” and does not account for the much greater spectrum of ethnicity, especially that seen in our ED population. Furthermore, their equations require that the provider performing calculations is aware of the patient’s exact age, which may not be possible in the unconscious, lethargic, or confused patient.

Secondly, opportunities for small confounders in the precision of obtaining accurate weight measurements arose in multiple steps of the data collection process. As some patient weights were obtained in less private and secure care areas, such as hallways and in triage, some patients were reluctant to remove articles of clothing such as pants and undershirts. Although we encouraged patients to

remove these articles of clothing whenever possible, some would not be dissuaded. We measured articles of clothing early in the study and found they had a negligible weight (b1 kg when weighed independently), but we recognize they may have made a small contribution that affected the actual weight measurement in the study. It is unclear whether this imprecision would tend to impact the use of the weight estimation tool and provider visual estimates to different degrees. Also, we relied on weights reported by inpatient nursing staff using calibrated bed scales for admitted patients rather than weights obtained using our own scales. The lack of calibration between these bed scales and ours, plus the possibility that the patients remained partially clothed, may have also impacted our ability to obtain precise measurements in admitted patients. Another interesting confounder we noticed was that the weight estimation method did not work as well in patients with certain comorbidities. For example, a few patients had large ascites secondary to end-stage liver disease. As would be expected, their weights exceeded that which was estimated by the formula. One could argue that this is not an unexpected observation and that patients with obvious physical examination findings that preclude use of a weight estimation tool should be excluded. However, we felt it was important to include all types of patients in our analysis, as the comorbid diagnoses of a patient presenting to an ED are not always obvious. Unfortunately, we did not have enough data involving patients with any specific comorbidities to make any statements about the validity of the weight estimation tool in these populations.

Thirdly, although we enrolled consecutive patients during the times we were present in the ED, most data collection was performed by the primary author between the hours of 9:00 AM to 5:00 PM. However, there is no reason to believe that patients’ weights would vary significantly during certain hours of the day.

The Ross Laboratories method is not the only available tool for weight estimation.

After an exhaustive search, we found multiple techniques using height, demi-span, and other physical parameters in calculating body mass index and ideal body weight [10-14]. In the study of Atiea [10], anthropometric parameters including midarm circumference, thigh circumference, knee height, waist skin fold thickness, and chest girth were used to create nomograms to estimate Geriatric patients‘ weights, then conventional multiple regression analyses were used to translate these into more easily usable equations. This study was limited in both its internal validity, as the authors acknowledge a positive bias when applying their equations to a prospective cohort in a second phase of their study, as well as its generalizability, as it was developed for Geriatric populations. Another study by Utley [11] used measurements of midarm circumference only and found a correlation between midarm circumference and actual body weight. However, they did not publish an attempt to formulate an equation to use measurements to precisely estimate weight,

and they acknowledged that their study included mostly young, healthy, white, ambulatory adults, thus limiting its generalizability. We had considered testing some of these weight estimation tools as well but ultimately decided that any technique that required too many steps or was too difficult to perform (ie, rolling a patient to measure chest girth) would not be easily adaptable for use by ED personnel and did not justify further testing.

In summary, an ideal method for bedside body weight estimation should be simple, rapid, and as accurate as possible. We believe that the use of the tool published by Ross Laboratories improves accuracy of weight estimation in ED populations and should be used in favor of visual estimations by health care providers when patients cannot be weighed or cannot state their own weight. Work on developing more accurate formulas should continue until such a time that all EDs are equipped with beds capable of weighing patients, given the potential for medication errors and the associated risks to patient safety.

References

1. Aspden P, Wolcott J, Bootman JL, Cronenwett LR, editors. Preventing medication errors: quality chasm series. Washington, DC: The National Academies Press; 2007.
2. Hall WL, Larkin GL, Trujillo MJ, et al. Errors in weight estimation in the emergency department: comparing performance by providers and patients. J Emerg Med 2004;27:219-24.
3. Anglemeyer BL, Hernandez C, Brice JH, Zou B. The accuracy of visual estimation of body weight in the ED. Am J Emerg Med 2004;22: 526-9.
4. Corbo J, Canter M, Grinberg D, Bijur P. Who should be estimating a patient’s weight in the emergency departmentp Acad Emerg Med 2005;12:262-6.
5. Kahn CA, Oman JA, Rudkin SE, et al. Can ED staff accurately estimate the weight of adult patientsp Am J Emerg Med 2007;25: 307-12.
6. The Ross Knee Height Caliper. 2004. bhttp://www.priorartdatabase. com/IPCOM/000006911/N Accessed 2008 Apr 1.
7. Streiner DL, Norman GR. Health Measurement Scales: a practical guide to their use and development. Oxford: Oxford University Press; 1989.
8. Weiner EA, Stewart BJ. Assessing individuals. Boston: Little Brown; 1984.
9. Menon S, Kelly AM. How accurate is weight estimation in the emergency departmentp Emerg Med Australas 2005;17(2):113-6.
10. Atiea JA, Haboubi NY, Hudson PR, Sastry BD. Body weight estimation of elderly patients by nomogram. J Am Geriatr Soc 1994; 42:763-5.
11. Utley R. Mid-arm circumference: estimating patients’ weight. Dim Crit

Care Nurs 1990;9(2):75-81.

1. Chumlea WC, Guo S, Roche AF, Steinbaugh ML. Prediction of body weight for the non-ambulatory elderly from anthropometry. J Am Diet Assoc 1988;88:564-8.
2. Donini LM, de Felice MR, de Bernardini L, Ferrari G, Rosano A, de Medici M, Cannella C. Body weight estimation in the Italian elderly. J Nutr Health Aging 1998;2(2):92-5.
3. Jung MY, Chan MS, Chow VSF, et al. Estimating geriatric patient’s body weight using the knee height caliper and mid-arm circumfe- rence in Hong Kong Chinese. Asia Pac J Clin Nutr 2004;13(3): 261-4.