Comparison of weight captured via electronic health record and cellular scales to the gold‐standard clinical method

Abstract

Introduction

Obtaining body weights remotely could improve feasibility of pragmatic trials. This investigation examined whether weights collected via cellular scale or electronic health record (EHR) correspond to gold standard in‐person study weights.

Methods

The agreement of paired weight measurements from cellular scales were compared to study scales from a weight loss intervention and EHR‐collected weights were compared to study scales from a weight loss maintenance intervention. Differential weight change estimates between intervention and control groups using both pragmatic methods were compared to study collected weight. In the Log2Lose feasibility weight loss trial, in‐person weights were collected bi‐weekly and compared to weights collected via cellular scales throughout the study period. In the MAINTAIN weight loss maintenance trial, in‐person weights were collected at baseline, 14, 26, 42 and 56 weeks. All available weights from the EHR during the study period were obtained.

Results

On average, in Log2Lose cellular scale weights were 0.6 kg (95% CI: −2.9, 2.2) lower than in‐person weights; in MAINTAIN, EHR weights were 2.8 kg (SE: −0.5, 6.0) higher than in‐person weights. Estimated weight change using pragmatic methods and study scales in both studies were in the same direction and of similar magnitude.

Conclusion

Both methods can be used as cost‐effective and real‐world surrogates within a tolerable variability for the gold‐standard.

Trial registration

NCT02691260; NCT01357551.

We compared the agreement of paired weight measurements from cellular scales versus study scales from a weight loss intervention and from the EHR versus study scales from a weight loss maintenance intervention. We also compared weight change estimates using both remote methods to study collected weight. Both methods can be used as cost‐effective and real‐world surrogates within a tolerable variability for the gold‐standard.

1. INTRODUCTION

The need to conduct pragmatic trials that maximize external validity and evaluate the feasibility of interventions in real‐world settings continues to grow. ¹ One strategy to decrease participant burden in such trials is to minimize in‐person study visits via remote data collection when possible. In addition to benefiting participants, remote data collection may result in less missing data and its associated bias and loss of power. Remote data collection may be particularly important for engaging populations with reduced resources, including time and accessibility to travel to research sites.

Weight can be collected pragmatically in various ways. In the past, self‐reported weight was the only option other than the in‐person study collected weight. ² , ³ , ⁴ More recently with technology improvements, objective strategies for collecting weight have been developed. One strategy is to obtain weights remotely through cellular or Wi‐Fi‐connected scales distributed to study participants. These scales offer the benefit of real‐time data collection when participants step on the scale and thus can be used to obtain more measurements than would be feasible with in‐person visits. Cellular scales are frequently used as part an intervention to promote self‐monitoring but are less often used to evaluate trial effects. ⁵ , ⁶ , ⁷ To date, no study to our knowledge has documented using the scales to reliably and validly assess weight change outcomes longitudinally. Previous work by Ross and others ⁸ compared cellular‐enabled BodyTrace™ scale weights to self‐reported weights. Another study by Pebley and others compared weights from cellular scales to in‐person collection at 3 time points out to 12 months; however, analyses focused on concordance and did not address the estimated effect size differences between the 2 methods. ⁹

A second strategy to pragmatic assessment of weight is to obtain electronic health record (EHR)‐collected weights from clinical visits, as researchers have done previously. ¹⁰ , ¹¹ , ¹² EHR abstraction avoids the financial cost of purchasing cellular or Wi‐Fi scales for each study participant. However, limitations include a lack of uniform protocol for collecting measurements (e.g., removal of heavy outer layers, shoes, etc.), bias due to lack of data for individuals who may not have clinical visits during the study period, or conversely, multiple repeated weights from individuals with many clinical visits. Recent work using EHR‐collected weights has primarily focused on improving accuracy and measurement and algorithms used for data extraction. ¹³ , ¹⁴ To our knowledge, only one set of analyses by Gallis and others has been conducted using EHR weights to examine intervention efficacy. ¹⁵ This work focused on weight loss and weight gain prevention and did not assess measurement concordance between study scale weights and weights collected via EHR. ¹⁵

In the research context, steps are taken to standardize the assessment of weight and to minimize variability and error in the measure. This may include removal of shoes and outerwear, including a fasting period before assessment, and clear instruction on times of day for each assessment of weight at baseline and follow up visits. Using cellular or Wi‐Fi connected scales, study teams may instruct participants about when and how to weigh. However, the extent to which participants follow these instructions cannot be verified, as others in the home could potentially stand on and use the scale. Use of these standards is also uncertain during clinical encounters when data is entered into the EHR. Because clinical trials of weight management interventions are longitudinal, a better understanding of the accuracy in estimating changes in weight is necessary for both EHR and cellular‐connected scales.

To address these gaps in knowledge and to illustrate the differences between both pragmatic weight data collection options, these two methods were compared to the gold standard in‐person study protocol weight collection in two different weight management trials: 1) Log2Lose weight loss feasibility study using cellular‐connected scales and 2) MAINTAIN weight loss maintenance study using EHR‐derived weights. The specific objectives of these analyses are to first assess the agreement of paired weight measurements between the two remote methods and study scale weights and then to compare estimated weight change effects longitudinally using the two remote methods and study scale weights. Practical issues to consider for both weight data capture methods are also discussed.

Study 1

Comparison of weight collection via cellular scales and study scales in the Log2Lose study

2. METHODS

2.1. Study overview

Log2Lose was a 25‐week trial that compared the feasibility of delivering different financial incentive strategies that might enhance weight loss. ¹⁶ Both study design and results are published. ¹⁶ , ¹⁷ Briefly, (N = 93) adults with obesity (BMI ≥ 30 kg/m²) were recruited in Durham, NC to a 4‐arm (2 × 2 factorial) behavioral weight loss intervention augmented with incentives for weekly weight loss, weekly dietary tracking, both, or neither.

All participants attended bi‐weekly classes for 24 weeks that focused on nutrition education and skill building. At the first meeting, study staff obtained a baseline weight using calibrated study scales (Tanita™, Arlington Heights, IL). Participants attended group sessions that took place on the same day of the week at the same time every 2 weeks in the afternoon or evening. Upon arrival at these sessions, participants were weighed on the study scale by study staff. They were instructed to wear light clothing and remove their shoes, heavy jewelry or pocket items and any heavy outer layers. ¹⁸ , ¹⁹

Cellular scales (Model BT004, BodyTrace™, New York, NY) were distributed at the first group session. Participants were instructed to weigh themselves a minimum of twice a week (preferably at the beginning and end of each week) to enable calculation of weekly weight change, but were encouraged to weigh themselves daily. They were advised to weigh in light clothing with shoes removed, first thing in the morning after voiding their bladder and before eating. Participants were asked not to let any other individuals in the house stand on the scale and to place the scale on a hard, flat surface.

Weights collected via cellular scales were used for intervention tracking and incentive distribution, while weights collected via study scales were used for outcome assessments. The cellular scale data‐‐including weights, dates and time stamps‐‐was collected into a digital health platform. Data cleaning as outlined in the protocol for the cellular scale data involved identifying implausible weights and excessive frequency of weights. First, the cellular scales had a maximum weight limit of 397 lbs., which was an eligibility criterion of 380 lbs. to account for possible weight gain; thus, there was no absolute upper weight limit algorithm applied for data cleaning purposes. Second, to minimize discrepancies due to different people standing on the scale, an algorithm built into the digital health platform invalidated any weight differing by greater than 10% from the prior weight in a single day. ⁵ If cellular scale weights were invalidated if they differed by greater than 20% from the baseline cellular scale weight. If more than one cellular scale weight was provided in a day, the first weight of the day was used. Finally, data cleaning that occurred after all study data were collected, involved identification and removal of observed changes greater than 10 pounds based on a computed average change per week between the two measures and visual inspection of weight trajectories to identify any large jumps or dips.

2.2. Analyses

2.2.1. Cellular scale and in‐person weight measurement agreement

All analyses were performed using the SAS/STAT component of SAS™ software (version 9.4, SAS Institute, Inc., Cary, NC). In‐person body weight measurements were paired with the closest home scale weight measurement within $\pm$ 3 days of the in‐person weight. The day the home scale weight was captured was then coded as to whether it occurred prior to, on, or after the day of the in‐person visit. For example, weights captured 3 days before the in‐person weight were coded −3 while weights captured 3 days after the in‐person weight were coded +3. Due to in‐person weight collection occurring every other week at the group sessions, each participant could contribute up to 13 in person weights that were paired with a home‐scale weight over the span of 25 weeks; for each, the difference between the cellular scale and in‐person measurement was calculated to assess concordance. Weight differences >4.5 kg (corresponding to 10 lbs.) were excluded as this was determined to be an unlikely amount of weight change in a 3‐day period.

The relationship between the two methods of body weight measurements was first examined using a Bland‐Altman (BA) plot. ²⁰ , ²¹ This helped to visualize the overall agreement of the two measurements by plotting the difference of the paired weights (Y‐axis) against the average of the pair (X‐axis), as well as lines representing the mean difference and 95% confidence interval limits.

A one‐sample t‐test of the difference between the two measurements followed by a linear regression model, predicting the in‐person weight from the cellular weight, were then conducted to test for bias and a departure from perfect concordance of the two measures. Mixed models were run in which cellular scale weights were regressed on study scale weights allowing in‐person correlation of multiple pairs of measures on the same person by including a random intercept for participant. Residual plots were used to examine variability and whether this variability changed as a function of study weight.

Finally, a linear regression model was employed to determine whether the magnitude of difference between the two measures (outcome) was due to the lag between the two measurements and whether the cellular scale measurement was before or after the in‐person weight.

2.2.2. Cellular scale and study weights estimation of intervention effects comparison

An overall measure of total weight change, the change from baseline to the final follow‐up visit, was computed as the estimated effect size. ²² The pair wise differences between the cellular and study scale were then compared across study arms. All cellular scale weights collected between the first in‐person visit, the point at which the scales were distributed, until week 25 corresponding to the final in‐person study weight data collection time point at 6 months were used in this analysis.

Two separate general linear mixed models were fit, one using the study scale weights and one using the cellular‐scale weights; model specification was identical for both. To estimate the effect size, a model assuming a common intercept, which constrains the baseline means to be equal across the four study arms due to randomization, ²³ was used to compare the differences in weight over time across the four treatment arms over the 25 weeks. Fixed effects included treatment group, linear and quadratic terms for time, and the interactions of treatment with the time effects. A person‐specific random intercept and slope were included to model within‐person correlation over time. The weight change estimated form the cellular scales was then compared to weight change estimated from the study scale.

3. RESULTS

3.1. Cellular scale and study scale weight agreement reliability

The cellular scales produced over 11,000 weight data points over the 25 weeks of measurements, resulting in an average of 4.7 weights each week per participant. Matching cellular scale data to study scale weights resulted in 649 pairs of body weight measures from n = 85 participants after 51 pairs were removed due to occurring outside the 3‐day window or having a greater than 4.5 kg discrepancy. Mean body weight was 103.5 ± 22.0 kg measured on the cellular scale within this window compared to 104.1 ± 21.9 kg measured on the study scales. The BA plot (Figure 1) shows the mean difference between cellular scale weights and in‐person weights was −0.6 kg (95% CI: −2.9, 2.2 kg).

FIGURE 1.

Bland‐Altman (BA) plot comparing cellular scale collected weight and in‐person weight in the Log2Lose study (n = 85)

The Shapiro‐Wilk statistic of 0.89 (p‐value <0.0001) indicated the differences in weights were not normally distributed. Additionally, more points on the BA plot above the upper limit indicated the distribution of differences between the two weights was right‐skewed; suggesting in‐person weights were higher than the home scale weights. Regression analyses suggest that cellular scale weights were, on average, lower than the study scale weights, and that this difference increased as weight increased (multiplicative bias of 0.01 kg: slope of 1.01 kg; SE: 0.002; p‐value <0.001 when testing slope = 1). The additive bias estimated cellular scale weight measurements (i.e., estimated intercept) were 1.3 kg (SE: 0.22; p < 0.001) lower than in‐person measurements. Subsequent models showed the cellular scale and in‐person weight measures did not differ significantly based on when the cellular scale weight was taken relative to the in‐person weight measure (−0.7 kg; SE: 0.107 p = 0.32).

3.2. Cellular scale and study scale estimation of intervention effects comparison

Table 1 shows the estimated average weight by treatment group at baseline and 25 weeks as well as weight change from baseline to 24 weeks using data collected from both methods from n = 91 participants. One participant withdrew from the study after the baseline visit and contributed no cellular‐scale weights. Another participant had significantly different cellular scale weights and in‐person weights flagged early on in the study. These two were excluded from analyses. To illustrate the difference between observations collected from both scales, the number of participants contributing to baseline and week 25 estimates is also given. There was no statistically significant treatment effect using either the cellular scale weight data or the in‐person weight data, which was not surprising given this was a small feasibility and not an adequately powered efficacy study. Weights at each time point were lower, and weight changes were smaller, for the cellular scale data than the in‐person data for all four groups (Table 1).

TABLE 1.

Cellular scale and in‐person model‐estimated weight change from baseline to 25 weeks in the Log2Lose study (n = 91)

Model‐estimated weight change from baseline to 25 weeks
	Group A		Group B		Group C		Group D
	Cellular weights	In‐person weights	Cellular weights	In‐person weights	Cellular weights	In‐person weights	Cellular weights	In‐person weights
Baseline; kg (95%CI)	106.6 (101.7, 111.5); n = 19	109.5 (105.1, 113.9); n = 23	106.6 (101.7, 111.5);n = 17	109.5 (105.1, 113.9); n = 21	106.6 (101.7, 111.5); n = 20	109.5 (105.1, 113.9); n = 23	106.6 (101.7, 111.5); n = 20	109.5 (105.1, 113.9); n = 24
25 weeks; kg (95%CI)	101.3 (95.3, 107.3); n = 21	102.9 (97.6, 108.3); n = 13	100.4 (94.2, 106.6);n = 18	102.9 (97.5, 108.4); n = 10	101.4 (95.4, 107.5); n = 15	103.9 (98.5, 109.3); n = 13	102.9 (96.7, 109.2); n = 15	104.5 (99.0, 110.1); n = 9
Weight change; kg (95%CI)	−5.3 (−8.8, −1.7)	−6.5 (−9.6, −3.5)	−6.2 (−10.0, −2.4)	−6.6 (−9.8, −3.4)	−5.2 (−8.7, −1.6)	−5.6 (−8.7, −2.5)	−3.7 (−7.6, 0.3)	−5.0 (−8.3, −1.6)

Note: Group A = incentivized for both weight loss and dietary tracking; Group B = incentivized for dietary tracking; Group C = incentivized for weight loss; Group D = no incentives.

Study 2

Comparison of weight collection via EHR and study scales in the MAINTAIN study

4. METHODS

4.1. Study overview

MAINTAIN was a two‐arm randomized controlled trial to test the effect of a weight loss maintenance intervention delivered via telephone. Patients (N = 222) of the Durham VA Medical Center who had obesity (BMI ≥ 30 kg/m²) were enrolled into a 16‐week run‐in weight loss program followed by randomization to maintenance intervention or usual care. Study design and results have been published. ²³ , ²⁴ Individuals were required to lose at least 4 kg in the weight loss program to be eligible for the maintenance phase. More information on this intervention has been published. ²³ Participants were veterans who received at least a portion of their medical care at the Durham VA. During the study, participants were weighed in person by trained study staff on calibrated study scales (Tanita™, Arlington Heights, IL) using the same protocol for weight collection described in Study 1 at baseline and weeks 14, 26, 42, and 56. All body weight measurements recorded from any outpatient clinical encounter at the VA during the corresponding time period were abstracted. Weight measurements taken from clinical encounters up to 8 weeks before study start were included to capture a baseline weight. Similarly, weights measured out to 8 weeks following study completion were included.

4.2. Analyses

4.2.1. Electronic health record and in‐person weight measurement agreement

Methods mirrored the agreement analysis used to compare cellular scales and study scale weights in Study 1. The analytic protocol stipulated that clinically implausible values over 700 lbs. and under 50 lbs. be omitted. Data were visually inspected to eliminate implausible weights defined as pairs with greater than a 15 kg difference in pairs collected 1 week from each other. For this analysis, only baseline in‐person weights were paired to weights abstracted from the EHR between 2 weeks prior and 2 weeks following the study scale weight collection. The concordance analysis relied on only paired baseline weights to reduce bias related to the EHR data capture process (i.e., since weight captured from the EHR is based on clinical encounters, those benefitting from weight loss and maintenance may have had fewer encounters, which could impact agreement between the two measures). Participants with one or fewer encounters in the EHR weighed less on average at enrollment in the maintenance program than those with multiple encounters (100 vs. 104 kg). These became more difficult to pair as clinical encounters occurred sporadically during the 56‐week follow‐up limiting this analysis to the baseline time point. As before, the relationship between the two methods of body weight measurements was examined visually, using a BA plot, ²⁰ , ²¹ followed by one‐sample t‐test to examine bias, and an associated 95% confidence interval on the observed bias computed. Linear models regressing the EHR weights on the study scale weights tested for a departure from perfect concordance of the two measures.

4.2.2. Estimated intervention effects comparison

As in Study 1, a general linear model with a constrained intercept was used. The outcome variable in this model was the study‐measured weight at weeks 0, 14, 26, 42, and 56. Initial weight loss stratum (<10 kg vs. ≥10 kg); indicator variables for weeks 14, 26, 42, and 56; and indicators for arm assignment interacted with each follow‐up time point indicator variable. Contrast statements were used to estimate the difference between the groups in weight regain from week 0 and each follow‐up. Results from the EHR model estimating change in weight from week 0 to week 56 across intervention and control groups were compared to the results of the general linear model used to estimate between group differences with study scale weights of the MAINTAIN trial. ²³

5. RESULTS

5.1. Electronic health record and in‐person weight agreement

Participants had an average of 6.1 weights (range 0–31) collected from 688 total clinical encounters during the 72‐week measurement period (56 weeks +/−8 weeks) abstracted from the EHR. In the 3‐week window surrounding the baseline visit, n = 60 participants had clinical encounters with a weight available in the EHR that could be paired to the in‐person baseline weight after excluding one pair of weight measurements due to a larger than 15 kg difference in the weights. Mean body weight measured in person was 105.7 ± 20.6 kg, compared to 108.5 ± 20.6 kg from the EHR. Electronic health record measurements were on average 6.6 (SD: 4.6) days from the in‐person visit.

The BA plot (Figure 2) showed the mean difference between EHR weights and in‐person weights was 2.8 kg (SE: −0.5, 6.0). The Shapiro‐Wilk statistic of 0.98 (p‐value = 0.55) indicated the differences in weights could be assumed normally distributed.

FIGURE 2.

Bland‐Altman (BA) plot comparing electronic health record (EHR) collected weight and in‐person weight in the MAINTAIN study (n = 60)

Regression analyses suggest that weights collected from the EHR were, on average, higher than the in‐person collected weights, and this difference was larger for people who weighed less (slope of 0.99 kg; SE: 0.01; p‐value <0.01). On average, EHR weight measurements calculated from the intercept were 3.0 kg (SE: 1.1; p = 0.01) higher than in‐person study scale measurements. Finally, models accounting for time from randomization and the difference between the EHR and in‐person weight did not reach statistical significance (−0.3 kg; SE: 0.4; p = 0.5), indicating that at baseline no statistical differences between whether the EHR visit was before or after the baseline visit.

5.2. Electronic health record and in‐person weight estimation of intervention effects comparison

Analyses using the study scale used weights from all n = 222 participants while analyses utilizing the EHR weights included only n = 211 participants due to 11 individuals not having a clinical encounter during the study period. Table 2 shows the estimated weight between the two treatment arms at baseline and 56 weeks using weight data collected from the in‐person study scale and the EHR. Weights at both time points were higher for the EHR estimates. However, weight changes were similar between the study scale weight and the EHR weight estimates (Table 2).

TABLE 2.

Electronic health record (EHR) and in‐person model‐estimated weight change from baseline to 56 weeks in the MAINTAIN study (n = 222)

Model‐estimated weight change out to 56 weeks (MAINTAIN; EHR)
	Intervention		Control
	EHR weights	In‐person weights	EHR weights	In‐person weights
Baseline; kg (95%CI)	106.6 (103.6, 109.6)	103.6 (100.9, 106.3)	106.6 (103.6, 109.6)	103.6 (100.9, 106.3)
56 weeks; kg (95%CI)	107.0 (103.7, 110.4)	104.3 (101.4, 107.2)	108.4 (105.1, 111.7)	105.9 (103.0, 108.9)
Estimated mean difference in differential weight change	1.3 kg (95%CI: −1.1, 3.7)		1.6 kg (95%CI: 0.07, 3.1)

6. DISCUSSION

Weights collected from in‐person study visits were compared to weights collected using two different pragmatic methods. While self‐reported weight data collection has shown good validity in large clinical ⁴ and non‐clinical samples ³ it is less precise in smaller samples. ²⁵ Weight captured from the EHR or cellular scales offer important improvements to researches conducting clinical trials. Our findings of the concordance and weight change comparisons suggest that both cellular scale and EHR collected weights yielded results that are comparable to the gold standard within an acceptable amount of error based on clinical interpretation. ²⁶

Concordance analyses assessing agreement between cellular scale and in‐person weights suggested that the cellular scale weights were slightly lower than the closest in‐person weight. The difference in the scale weights appears plausible given that participants were instructed to weigh themselves at home first thing in the morning in light clothing, and in‐person weights were taken at evening and afternoon class sessions with participants wearing light clothing. Our analyses indicated that heavier participants had an increased discrepancy between cellular scale weight and study scale weight. This may be due to people with more weight experiencing greater weight fluctuations, or that people with larger bodies have more, and thus heavier clothing. The multiplicative bias may not be clinically important as it implies that two individuals who differ in weight by 10 kg would have an additional discrepancy between in‐person and cellular weights of one‐10th of a kilogram.

Our finding of high agreement overall between the cellular scale weights and in‐person weights is similar to and extends the findings by Pebley and others using data from smart scales and in‐person visits over a 12‐month period. ⁹ While Pebley and others examined concordance of paired weights collected on the same day and those paired out to 7 days after, our analyses of whether the cellular scale weight was taken on the days prior to or following the in‐person weight measure suggested that a 3‐day window before or after the in‐person visit is acceptable for data collection.

Results of the weight change comparison highlighted differences in the weight change estimates of the control group. These differences may be due to fewer participant weights captured from the study scale (n = 9 from the study scale vs. n = 15 from the cellular scale as reported in Table 1). As the cellular weights were able to capture more weights from individuals who may have stopped attending classes, or were lost to follow‐up, this difference highlights the potential opportunity to overcome biases due to missing data for individuals who stopped attending sessions. Future work should utilize cellular scale data not only to inform missing data but also to examine patterns to better understand and strategize how to prevent disengagement. The much larger number of weights obtained from the cellular scales than the in‐person study scales is also important when calculating power for an appropriate effect size, and may allow for smaller sample sizes.

Further work is needed to fully understand how frequently cellular scales may yield valid data beyond a 25‐week weight loss trial. Following the first 6 months of weight loss, individuals in interventions tend to experience weight regain and weigh less often. While Pebley and others examined agreement out to 12 months, more work is needed to understand longitudinal trajectories in self‐monitoring. These scales could be used to collect data beyond 6 months to inform sustainment of weight loss maintenance and characterizing the timing of declines in self‐monitoring.

Other implications related to study processes of these scales include instructing participants on use. This differs from the clinical weights in the EHR, where less instruction could be given to weighing. However, given that home scales carry the risk of multiple individuals in the home providing weights, the certainty of capturing the weight from the correct individual was higher for the EHR. While instructions for weighing at home including weighing on a hard flat surface and instructing participants that no other individuals in the home should use the scale resulted in fairly reliable data, strategies to identify and clean multiple and implausible weights were still necessary.

Unlike the cellular scale results, BA plots and regression modeling suggested the EHR collected weights were higher than those collected in person, consistent with the work of Gallis and others. ¹⁵ This is likely due to the lack of protocol for collecting weights in a clinical setting; individuals may be weighed with heavy outer layers or jewelry, items in pockets and while wearing shoes. In some clinical settings patients are asked to self‐report their current weight instead of being weighed potentially adding further error to clinical weight collection in the EHR. Additionally, larger time windows are needed to capture enough EHR weights from clinical encounters to compare agreement with study weights, which may result in greater discrepancies between EHR weights and in‐person study weights.

Results of the effect size comparison of EHR collected weight data to in‐person collected weight data suggested that the estimated intervention effects on weight regain were quite similar between methods. Thus, supporting the use of EHR weight collection for long‐term weight management studies, in which the study time period is long enough to capture enough weights from clinical encounters. Indeed, EHR weight data has been used to report lasting intervention effects following the conclusion of the clinical trial. ²⁷ Future work focusing on strategies to increase uniformity of weight collection from clinical encounters is needed to inform pragmatic trials as well as system‐level weight interventions.

7. CONCLUSION

As trials move toward pragmatic methods to aid in implementation and translation to clinical practice, an understanding is needed of alternative methods to the gold‐standard in‐person collection of weight data. The COVID‐19 pandemic has led to more medical care and research shifting to remote and telehealth practice. More patients have now experienced remote care and are interested in continuing to engage in telemedicine and remote visits. ²⁸ , ²⁹ , ³⁰ Weight loss and maintenance trials can consider both cellular or Wi‐Fi enabled home scales and the EHR to estimate weight change as cost‐effective and real‐world surrogates for the gold standard. While each method requires an understanding of the potential sources of error and the implications of the data‐generating process, both methods offer benefits including decreased participant burden and potentially easing missing data due to participant loss to follow‐up. Moving forward, work is needed to better understand how these methods may also contribute to improving estimation when missing data is present. Additionally, studies that are already utilizing these methods along with in‐person weight collection should report findings from each method in order to build a larger body understanding.

AUTHOR CONTRIBUTIONS

Kara L. Gavin, performed analyses and interpreted the results and wrote the manuscript; Emily J. Almeida contributed to analyses and interpreted results and reviewed and provided edits for the manuscript; Corrine I. Voils provided content expertize for interpreting the data, provided revisions and edits to the manuscript; Melissa M. Crane provided content expertize for interpreting the data, provided revisions and edits to the manuscript; Ryan Shaw provided content expertize for interpreting the data, provided revisions and edits to the manuscript William S. Yancy provided content expertize for interpreting the data, provided revisions and edits to the manuscript; Jane Pendergast led analyses, and provided edits and revisions to the manuscript; Maren K. Olsen led analyses, and provided edits and revisions to the manuscript; All authors reviewed and approved the final version of the manuscript.

CONFLICT OF INTEREST

We have no conflict of interest to disclose.

Gavin KL, Almeida EJ, Voils CI, et al. Comparison of weight captured via electronic health record and cellular scales to the gold‐standard clinical method. Obes Sci Pract. 2023;9(4):337‐345. 10.1002/osp4.656