Cardiometabolic Characteristics of Weight Cycling: Results from a Mid-South Regional Comprehensive Healthcare System

Abstract

Objective:

To determine the unique clinical and cardiometabolic risk characteristics of weight cyclers and identify differences between weight cyclers and individuals with other weight change trajectories.

Methods:

A de-identified database of 1,428,204 Vanderbilt University Medical Center patients from 1997–2020 was included based on having ≥5 years of recorded weights. Patients with history of malignant neoplasm, bariatric surgery, implausible BMI (<15, >80) or missing documented height were excluded yielding 83,261 participants categorized by weight trajectory: weight-stable, weight-gainer, weight-loser, or weight-cycler based on criteria of ≥5% weight-change thresholds. Additionally, quartiles of average successive weight variability (ASV) were evaluated to determine the effect of absolute differences between successive weight values.

Results:

Over half (55%) were weight-cyclers, 23% weight-gainers, 12% weight-losers, 10% weight-stable over 5 years. Although baseline BMI did not differ among groups, weight-cyclers were more likely to have lower HDL-cholesterol and higher blood glucose and triglycerides levels, and have been prescribed antihypertensive, dyslipidemia, and/or anti-diabetic therapies. They were also younger and more likely to be smokers. Participants with the greatest weight variability (highest quartile of ASV) had higher cardiometabolic risk scores.

Conclusions:

Weight cycling was highly prevalent yet yielded no meaningful overall change in body weight after 5 years. These findings support a paradigm shift in weight management in persons with overweight/obesity toward reducing cardiometabolic risk with or without weight loss.

1. Introduction

The prevalence of excess body weight has risen to an epidemic level in the U.S. with 1 in 3 adults now overweight and 2 in 5 adults having obesity based on body mass index (BMI). Although relationships between obesity and adverse health outcomes, including type 2 diabetes (T2DM), are well established,^1,2 50–80% of individuals who lose weight gain it back within 5 years.^3–5 Repeated attempts to manage weight that result in successive bouts of loss followed by gain, termed yo-yo dieting or weight cycling, are also associated with deleterious health consequences from T2DM to increased mortality.^6,7 Though studies show mixed results,^8–12 two recent meta-analyses indicate greater pooled risk from weight cycling for incident T2DM, cardiovascular disease, and all-cause mortality.^7,13

A possible mechanism linking weight cycling to cardiometabolic risk is via pro-inflammatory changes in adipose tissue, with infiltration of macrophages, T-cells, and other immune cells that secrete signaling molecules including adipokines and pro-inflammatory cytokines.¹⁴ While weight loss may reduce adipose inflammation, quiescent pro-inflammatory cells remain in tissues, and weight regain may increase numbers and sensitivity of immune cells exacerbating cytokine production in the setting of weight cycling.^15,16 Further indicating a role for immune cell adaptations in cardiometabolic dysfunction, weight-cycled mice have impaired glucose tolerance and adipose tissue insulin sensitivity.¹⁷ In humans, whether weight cycling has similar adverse effects remains unclear.

Inferring conclusions about risks and outcomes from weight cycling in humans is challenged by lack of a consensus definition and limited availability of large population-based cohorts and longitudinal datasets. Published definitions range from one or more episodes of 5% weight loss and regain,⁶ 10lb loss and regain,¹² the degree of variation from weight to weight,¹⁸ and self-report of any amount of loss and regain.¹⁹ Using the criteria of ≥5% weight loss and gain, a large study in England employing electronic health record (EHR) data identified 50% of patients with obesity as weight cyclers.⁵ Two smaller cohort studies in the U.S. estimated 27–50% of adults have a weight history with at least one episode of cycling.^12,20

Given the potential for serious cardiometabolic health impact and the paucity of investigation on weight trajectories, the present study aimed to determine the prevalence of weight related trajectories and their unique characteristics in a mid-South population from a regional medical center comprised of 7 hospitals and >180 outpatient clinics across Tennessee and neighboring states. We hypothesized that a history of weight cycling and/or high weight variability, compared to overall weight stability, weight loss, or weight gain would have worse clinical and cardiometabolic risk profiles. This information can provide insights for optimizing clinical weight management practices.

2. Methods

2.1. Eligibility

A de-identified dataset including demographics, ICD9/10 diagnoses, laboratory values, medications, smoking status, weights, heights, and BMIs for all Vanderbilt University Medical Center (VUMC) patients from 1997 through 2020 was extracted from the EHR. Patients were included if they were >18 years, had ≥5 years of successive weights documented, and each weight ≤18 months from prior weight. Patients were excluded if there was history of bariatric surgery, malignant neoplasm other than non-melanoma skin cancer (ICD-9 140–210, excluding 173; ICD-10 C00–96, excluding C44), no recorded height, or any documented BMI was implausible (<15 or >80 kg/m²), resulting in a cohort of 83,261 adults (Figure 1). The study was exempt by the VUMC Institutional Review Board (IRB #230831). We utilized the STROBE (Strengthening Reporting of Observational Studies in Epidemiology) cohort studies reporting guideline.

Figure 1:

Flow Diagram of Study Inclusion and Exclusion Criteria

2.2. Dataset Cleaning

Weights and heights were cleaned according to a pre-specified algorithm. Heights were converted to centimeters with the assumption that recorded height measurement of 0.9–2.3 was meters, 3.0–7.5 was feet, and 36.0–89.9 was inches. Height diverging >3% from a participant’s median was replaced with the nearest plausible height, unless there was diagnosis associated with significant height change (osteoporosis, spinal stenosis, amputation). Weights were cleaned in multiple steps. First, weights recorded during pregnancy were omitted (90 days before or 270 days after positive β-HCG test or any pregnancy-related ICD 9/10 code). Next, if weight was over 1.5 times a participant’s median weight it was assumed to be measured in pounds and converted to kg. Any weights considered physiologically implausible (≤ 30 or ≥ 250kg) were removed unless there was a weight-related diagnosis (with dwarfism or anorexia the threshold was lowered to ≤ 20kg and with diagnosis of extreme weight gain, severe obesity, or with ≥ 3 weights > 250kg, the upper threshold was increased to ≥ 450kg). Weights varying more than 33% of an individual’s median within 12 months, 20% over 60 days, 14% over 30 days, or 12% over 21 days were omitted. Because the dataset included in- and outpatients, it was necessary to reduce bias from short-term weight fluctuations (e.g., daily weights during hospitalization) and create a standardized number of datapoints. Thus, weights were linearly interpolated to a 60-day interval creating a grid of weights for each participant with one weight measurement every 60 days over the 5-year observation period. Any periods with no recorded weight within 90 days were left missing. Repeat weights on the same day were also omitted from analysis. Linear interpolation was performed using R version 4.3.2. (R Foundation for Statistical Computing, Vienna, Austria). BMI was calculated from the cleaned weights and median height.

2.3. Weight Patterns and Trajectories

Total weight change was calculated as last recorded weight during the 5-year period minus baseline weight. Maximum weight change was calculated as difference between maximum and minimum weights during this timeframe. Average successive weight variability (ASV), a measure of weight-to-weight fluctuation, was calculated from the interpolated grid as $\mathrm{A}\mathrm{S}\mathrm{V}=\frac{\sum _1^{n-1}|{\mathrm{W}}_{n+1}-{\mathrm{W}}_n|}{n-1}$, with W being weights in the interpolated grid (skipping blank values) and n being total number of positions on the interpolated grid (n=30).²¹ ASV is often divided into quartiles with the highest quartile identifying weight cycling as high variability indicates many fluctuations.²¹ Participants were categorized as weight-stable, weight-gainer, weight-loser, or weight-cycler using the interpolated grid and a clinically significant threshold of ≥5% change²² via a program developed in Python 3.10.12 (Python Software Foundation, Beaverton, Oregon) which sequentially compared weights to identify when a weight value differed by ≥5% from the previous extrema (maximum or minimum). Thus, the 5% threshold from any prior weight was employed rather than merely calculating 5% change from baseline. Participants categorized as weight-stable never had two weights differing ≥5%. Participants categorized as weight-gainers had at least one period of ≥5% gain without subsequent loss. Participants categorized as weight-losers had at least one episode of ≥5% weight loss without subsequent regain. Weight-cyclers were defined as having ≥5% weight change from a local extrema and subsequent change from the new extrema in the opposite direction. Weight cycle length is time in which a participant experienced a complete bout meeting the ≥5% criteria for loss and regain. Average weight change per cycle is the absolute sum of total weight gained (minimum to maximum) and total weight lost (maximum to minimum) over the weight cycle course, averaged across all cycles. Weight trajectories were also calculated using a cut-point of 10% for comparison to 5% (findings reported are 5% cut-point unless stated otherwise).

2.4. Cardiometabolic Risk

Laboratory-based risk factors (glucose, creatinine, estimated glomerular filtration rate (eGFR), total cholesterol, LDL-cholesterol, HDL-cholesterol, and triglycerides) were extracted from the EHR. The value closest to the date of baseline weight (within 2 years) was included in the dataset. Participants with no laboratory values meeting this criterion ranged from 20% for glucose levels to 50% for lipid levels. Values for possibly pregnant participants were excluded. An overall cardiometabolic risk score, adapted from Grundy et al,²³ was created using laboratory and medication data. Scores ranging from 0–5 points were calculated with 1 point for waist circumference ≥102cm in males and ≥88cm in females (BMI ≥25 was a proxy for participants with no waist circumference²⁴); 1 point for triglycerides ≥150mg/dL and 1 point for HDL <40 in males or <50 in females (or 2 points if the participant was on anti-hyperlipidemic therapy); 1 point for antihypertensive therapy; and 1 point for random glucose ≥140mg/dL²⁵ or antidiabetic therapy.

2.5. Statistical Analysis

Data are presented as means with standard deviations or frequency with numbers and percentages. Normality of data was determined using the Shapiro-Wilk test and observation of histograms. Statistical analyses were performed using R version 4.3.2. with Type I error of α < 0.05 considered significant. Effect sizes quantified the magnitude of significant results since p values are affected by sample size. For continuous variables, the Kruskal-Wallis was calculated with effect size (ε²: 0.01–0.06 small, 0.06–0.14 moderate, ≥0.14 large effect)²⁶ followed with post-hoc Dunn test employing Benjamini-Hochberg correction. For between-sex differences in measures of weight variability among cyclers, the Mann-Whitney U test was calculated. For categorical variables, including sex, age and BMI differences, the Pearson’s Chi-Squared test was calculated with effect size (Cramer’s V: ≤0.1 weak, 0.3 moderate, ≥0.5 strong association).²⁷

3. Results

3.1. Descriptives and ASV

Of 83,261 participants, 60% were female (Table 1), which resembled the proportion in the full VUMC EHR (58% female). Average age at baseline was 51.3 ± 15.0 years for males and 49.0 ± 15.8 years for females. Over the 5-year observation period, males had a median of 14 (10, 21) visits and females had 15 (10, 22) visits with recorded weights. The average BMI was 29.2 ± 5.7 kg/m² for males and 28.4 ± 7.1 kg/m² for females, with 40% of males and 29% of females meeting criteria for overweight and 38% of males and 34% of females meeting criteria for obesity. Mean ASV was 0.88 ± 0.60 kg in males and 0.80 ± 0.54 kg in females. Participants in the highest quartile of ASV were younger and had higher BMIs (Table 2). They were more likely to have prescribed antihypertensive and antidiabetic therapies (ε² ≥2.0, Table 3). Average cardiometabolic risk score ranged from 1.89 ± 1.28 in the lowest quartile of ASV to 3.23 ± 1.31 in the highest quartile, with a moderate effect size (ε² 0.07) for the mean difference among quartiles. Average ASV was strongly associated with weight trajectory category in males (ε² 0.48) and females (ε² 0.40). Categorizing into weight trajectories yielded 12% of males as weight-stable, 14% weight-losers, 23% weight-gainers, and 50% weight-cyclers. In females, 8% were weight-stable, 13% weight-losers, 23% weight-gainers, and 57% weight-cyclers (Figure 2). Mean ASV was significantly lower in the weight-stable than other groups at 0.34 ± 0.17 kg and 0.27 ± 0.14 kg in males and females, respectively. In contrast, ASV was 0.61 ± 0.27 kg in males and 0.53 ± 0.27 kg in females among weight-losers, 0.63 ± 0.29 kg in males and 0.54 ± 0.27 kg in females among weight-gainers, and 1.20 ± 0.65 kg in males and 1.03 ± 0.58 kg in females among weight-cyclers. Weight-cyclers had ~2-fold greater ASV than weight-losers or weight-gainers and a 3-fold greater ASV than weight-stable participants (p<0.0001).

Table 1a.

Baseline Demographics by Weight Trajectory Group in Males

	All	Stable	Gain	Loss	Cycle	P-value	Effect Size
Count (n)	33,060 (100%)	4,075 (12%)	7,631 (23%)	4,695 (14%)	16,659 (50%)
Age (years)	51.25 ± 15.04	54.54 ± 13.18a	48.52 ± 14.80b	56.52 ± 14.42c	50.21 ± 15.24d	2.24E-227	Є2 = 0.03
Height (cm)	178.79 ± 7.40	179.16 ± 7.09a	178.95 ± 7.29ab	178.68 ± 7.24bc	178.65 ± 7.57c	8.62E-04	Є2 < 0.01
Weight (kg)	93.58 ± 20.05	91.79 ± 16.66a	91.51 ± 18.74a	95.35 ± 19.16b	94.46 ± 21.48c	1.98E-37	Є2 < 0.01
BMI (kg/m2)	29.21 ± 5.70	28.55 ± 4.63a	28.53 ± 5.39a	29.82 ± 5.48b	29.52 ± 6.08c	2.69E-56	Є2 < 0.01
BMI Category						1.73E-99	V = 0.07
Underweight	250 (1%)	13 (0%)	63 (1%)	10 (0%)	164 (1%)
Normal Weight	6,983 (21%)	830 (20%)	1,860 (24%)	758 (16%)	3,535 (21%)
Overweight	13,389 (40%)	1,993 (49%)	3,192 (42%)	2,001 (43%)	6,203 (37%)
Class I Obesity	7,995 (24%)	920 (23%)	1,701 (22%)	1,241 (26%)	4,133 (25%)
Class II Obesity	2,929 (9%)	236 (6%)	552 (7%)	470 (10%)	1,671 (10%)
Class III Obesity	1,514 (5%)	83 (2%)	263 (3%)	215 (5%)	953 (6%)
Race (self-reported)						1.90E-23	V = 0.04
White	29,394 (89%)	3,718 (91%)	6,818 (89%)	4,256 (91%)	14,602 (88%)
Black	2,815 (9%)	238 (6%)	595 (8%)	311 (7%)	1,671 (10%)
Other	851 (3%)	119 (3%)	218 (3%)	128 (3%)	386 (2%)
Smoking Status						8.00E-47	V = 0.06
Non-Smoker	17,509 (53%)	2,443 (60%)	4,356 (57%)	2,428 (52%)	8,282 (50%)
Ever-Smoker	12,656 (38%)	1,356 (33%)	2,746 (36%)	1,845 (39%)	6,709 (40%)
Unknown	2,895 (9%)	276 (7%)	529 (7%)	422 (9%)	1,668 (10%)

Table 2a.

Baseline Demographics by Weight Variability (ASV) Quartiles in Males

	All	Q1	Q2	Q3	Q4	P-value	Effect Size
Count (n)	33,060 (100%)	8,265 (25.0%)	8,265 (25.0%)	8,265 (25.0%)	8,265 (25.0%)
Age (years)	51.25 ± 15.04	52.64 ± 14.72a	51.84 ± 15.11b	50.91 ± 15.32c	49.60 ± 14.82d	1.61E-36	Є2 < 0.01
Height (cm)	178.79 ± 7.40	177.91 ± 7.15a	178.52 ± 7.30b	178.89 ± 7.47c	179.84 ± 7.55d	8.08E-64	Є2 < 0.01
Weight (kg)	93.58 ± 20.05	86.08 ± 14.41a	90.41 ± 16.66b	94.76 ± 18.66c	103.07 ± 24.85d	<0.0001	Є2 = 0.09
BMI (kg/m2)	29.21 ± 5.70	27.15 ± 4.01a	28.33 ± 4.73b	29.57 ± 5.34c	31.80 ± 7.16d	<0.0001	Є2 = 0.08
BMI Category						<0.0001	V = 0.19
Underweight	250 (1%)	65 (1%)	67 (1%)	54 (1%)	64 (1%)
Normal Weight	6,983 (21%)	2,454 (30%)	1,867 (23%)	1,469 (18%)	1,193 (14%)
Overweight	13,389 (40%)	4,040 (49%)	3,734 (45%)	3,227 (39%)	2,388 (29%)
Class I Obesity	7,995 (24%)	1,387 (17%)	1,924 (23%)	2,339 (28%)	2,345 (28%)
Class II Obesity	2,929 (9%)	275 (3%)	518 (6%)	856 (10%)	1,280 (15%)
Class III Obesity	1,514 (5%)	44 (1%)	155 (2%)	320 (4%)	995 (12%)
Weight Group						<0.0001	V = 0.41
Stable	4,075 (12%)	3,322 (40%)	642 (8%)	107 (1%)	4 (0%)
Gainer	7,631 (23%)	2,655 (32%)	2,805 (34%)	1,646 (20%)	525 (6%)
Loser	4,695 (14%)	1,692 (20%)	1,821 (22%)	947 (11%)	235 (3%)
Cycler	16,659 (50%)	596 (7%)	2,997 (36%)	5,565 (67%)	7,501 (91%)
Race (self-reported)						9.706E-91	V = 0.08
White	29,394 (89%)	7,503 (91%)	7,456 (90%)	7,356 (89%)	7,079 (86%)
Black	2,815 (9%)	430 (5%)	583 (7%)	744 (9%)	1,058 (13%)
Other	851 (3%)	332 (4%)	226 (3%)	165 (2%)	128 (2%)
Smoking Status						4.406E-71	V = 0.07
Non-Smoker	17,509 (53%)	4,945 (60%)	4,502 (54%)	4,221 (51%)	3,841 (46%)
Ever-Smoker	12,656 (38%)	2,774 (34%)	3,085 (37%)	3,299 (40%)	3,498 (42%)
Unknown	2,895 (9%)	546 (7%)	678 (8%)	745 (9%)	926 (11%)

Table 3a.

Clinical Biomarkers and Prescribed Cardiometabolic Medications by Weight Variability (ASV) Quartiles in Males

	All	Q1	Q2	Q3	Q4	P-value	Effect Size
Count (n)	33,060 (100%)	8,265 (25.0%)	8,265 (25.0%)	8,265 (25.0%)	8,265 (25.0%)
Medications
Antihypertensive Therapy	14,229 (43%)	2,416 (29%)	3,123 (38%)	3,736 (45%)	4,954 (60%)	<0.0001	V = 0.25
Dyslipidemia Therapy	6,603 (20%)	1,256 (15%)	1,567 (19%)	1,697 (21%)	2,083 (25%)	1.93E-57	V = 0.09
Antidiabetic Therapy	9,940 (30%)	1,316 (16%)	2,043 (25%)	2,702 (33%)	3,879 (47%)	<0.0001	V = 0.23
Clinical Biomarkers
Glucose (mg/dL)	108.71 ± 47.58	101.31 ± 32.01a	106.48 ± 42.32b	110.43 ± 49.05c	116.01 ± 60.13d	5.9924E-52	Є2 < 0.01
Createnine (mg/dL)	1.14 ± 0.95	1.03 ± 0.38a	1.06 ± 0.53a	1.13 ± 0.90a	1.34 ± 1.51b	4.75E-3	Є2 < 0.01
EGFR (mL/min)	85.57 ± 24.28	85.64 ± 19.86a	85.87 ± 21.67ab	86.32 ± 24.60b	84.49 ± 29.49ab	1.33E-02	Є2 < 0.01
Total Cholesterol	180.83 ± 45.95	183.04 ± 42.12a	182.75 ± 43.96a	180.70 ± 47.14b	177.08 ± 49.64c	9.3113E-23	Є2 < 0.01
LDL-Cholesterol	102.23 ± 41.77	107.46 ± 38.15a	104.88 ± 40.49b	100.81 ± 42.51c	96.28 ± 44.50d	1.42E-59	Є2 < 0.01
HDL-Cholesterol	44.75 ± 14.15	47.60 ± 14.25a	45.56 ± 13.95b	44.03 ± 13.78c	42.04 ± 14.05d	1.50E-138	Є2 = 0.02
Triglycerides	155.41 ± 114.31	135.73 ± 91.87a	150.64 ± 106.83b	161.76 ± 122.71c	171.91 ± 127.6d	8.6827E-81	Є2 = 0.01
Cardiometabolic Risk Score	2.58 ± 1.40	1.89 ± 1.28a	2.39 ± 1.33b	2.72 ± 1.33c	3.23 ± 1.31d	<0.0001	Є2 = 0.07

Figure 2:

Proportion of Participants in Weight Change Trajectory Groups Using 5% and 10% Cut-Points

3.2. Weight-Stable Group

Significantly more males were weight-stable compared to females with weight change ≥5% (males: 12.3%, females: 7.6%, p=1.53e-147) and ≥10% (males: 53.3%, females: 40.9%, p=1.8e-271). Among all, the proportion that were weight-stable increased over age quartiles, with 7.6% of males and 4.2% of females aged 18–40 years vs 14.5% of males and 10.3% of females aged 60–85 years weight-stable (males: p= 2.11E-26, females: p= 2.39E-66). Weight-stable males were more likely to have overweight and less likely to have Class III obesity compared to other trajectories. Weight-stable females were more likely to have normal weight and less likely to have any BMI obesity class. Thus, males with overweight or obesity were twice as likely as females with overweight or obesity to be weight-stable (15% vs 7%, p= 7.80E-104, Cramer’s V 0.13 and 10% vs 6%, p= 9.74E-44, Cramer’s V 0.08).

Weight-stable participants were more likely to be non-smokers (60% of males and 75% of females, p=8.00E-47) and less likely to have prescribed antihypertensive (males: 31%, p=8.44E-146, Cramer’s V 0.14; females: 27%, p=8.33E-139, Cramer’s V 0.11), dyslipidemia (males: 17%, p=1.74E-20, Cramer’s V 0.05; females: 8%, p=3.56E-42, Cramer’s V 0.06) or antidiabetic therapies (males: 17%, p=1.69E-155, Cramer’s V 0.15; females: 11%, p=6.98E-187, Cramer’s V 0.13) (Table 4). Although overall effects were small (ε² <0.01), weight-stable participants had significantly lower triglycerides (males: p=2.40E-12, females: p=6.68E-47) and higher HDL-cholesterol (males: p= 4.76E-30, females: p= 1.02E-73) than other groups. LDL-cholesterol levels were not significantly different between weight-stable and weight-gain groups, however, weight-stable males had higher LDL than weight-loser or weight-cycler males (p= 1.31E-07 and p= 1.94E-14, respectively). Glucose levels also did not differ between weight-stable and weight-gain groups, but glucose levels were lower in the weight-stable group than weight-cyclers (males: p=2.58E-06, females: p=9.15E-04) and weight-losers (males: p=1.10E-13, females: p=4.34E-20).

Table 4a.

Clinical Biomarkers and Prescribed Cardiometabolic Medications by Weight Trajectory Group in Males

	All	Stable	Gain	Loss	Cycle	P-value	Effect Size
Count (n)	33,060 (100%)	4,075 (12%)	7,631 (23%)	4,695 (14%)	16,659 (50%)
Medications
Antihypertensive Therapy	14,229 (43%)	1,259 (31%)	2,786 (37%)	1,928 (41%)	8,256 (50%)	8.44E-146	V = 0.14
Dyslipidemia Therapy	6,603 (20%)	683 (17%)	1,317 (17%)	1,002 (21%)	3,601 (22%)	1.74E-20	V = 0.05
Antidiabetic Therapy	9,940 (30%)	707 (17%)	1,810 (24%)	1,467 (31%)	5,956 (36%)	1.69E-155	V = 0.15
Clinical Biomarkers
Glucose (mg/dL)	108.71 ± 47.58	102.64 ± 31.64a	105.23 ± 41.86a	110.45 ± 44.26b	111.22 ± 53.40c	3.15E-29	Є2 < 0.01
Createnine (mg/dL)	1.14 ± 0.95	1.05 ± 0.32a	1.08 ± 0.74b	1.08 ± 0.53a	1.21 ± 1.19c	0.000000106	Є2 < 0.01
EGFR (mL/min)	85.57 ± 24.28	83.55 ± 19.13a	87.77 ± 22.44b	83.18 ± 21.80a	85.72 ± 26.59c	1.40E-39	Є2 < 0.01
Total Cholesterol	180.83 ± 45.95	183.11 ± 41.76a	182.98 ± 44.51a	179.52 ± 45.99b	179.66 ± 47.48b	6.12E-10	Є2 < 0.01
LDL-Cholesterol	102.23 ± 41.77	106.43 ± 38.22a	105.60 ± 40.77a	101.39 ± 41.72b	99.89 ± 42.87b	1.55E-24	Є2 < 0.01
HDL-Cholesterol	44.75 ± 14.15	46.76 ± 13.92a	45.39 ± 13.74b	44.59 ± 13.84c	44.02 ± 14.42d	4.76E-30	Є2 < 0.01
Triglycerides	155.41 ± 114.31	142.18 ± 95.41a	149.93 ± 104.92b	156.72 ± 120.31b	160.77 ± 120.40c	2.41E-12	Є2 < 0.01
Cardiometabolic Risk Score	2.58 ± 1.4	2.09 ± 1.29a	2.33 ± 1.36b	2.61 ± 1.36c	2.80 ± 1.41d	3.80E-144	Є2 = 0.02

3.3. Weight-Loser Group

Among all participants, 14% of males and 13% of females had weight loss of ≥5% over their 5-year period. In the weight-loser group, 30% of males and 42% of females also met criteria for ≥10% weight loss. Weight-losers were significantly older than other participants (males: 6.2 years, p=2.24E-227; females 7.3 years, p<0.00001). The prevalence of weight loss increased across age quartiles (9% of males and 7% of females aged 18–39.9, 11% of males and 10% of females aged 40–49.9, 14% of males and females aged 50–59.9, and 15% of males and 20% of females aged 60–85 years). Baseline BMI did not differ when comparing weight-losers to weight-stable or weight-gainer groups and the prevalence of weight loss was similar in those with overweight or obesity regardless of sex. However, the amount of total weight loss increased steadily across BMI category from −3.85 ± 3.37 kg to −13.14 ± 10.61 kg (Figure 4). Weight-losers had higher baseline glucose and triglyceride levels, lower HDL, more prescriptions for anti-hypertensive, dyslipidemia, and antidiabetic therapies, and overall worse cardiometabolic risk score than weight-stable and weight-gainer participants. However, the magnitude of these effects was small in males and females.

Figure 4a-d:

Total and Maximum Weight Change by Weight Trajectory Group Stratified by BMI Category

3.4. Weight-Gainer Group

Of 19,099 participants who met criteria for ≥5% weight gain, 38% of males and 50% of females met criteria for ≥10% weight gain. Those with weight gain were significantly younger compared to other groups (males −3.55 years, p=2.24E-227; females −3.18 years p<0.00001). No difference was observed in baseline BMI between weight-gainer and weight-stable groups. Both total and maximum weight gain were similar between males and females, with total weight gain averaging 6.9 ± 6.5kg in males and 7.1 ± 6.3kg in females and maximum weight gain averaging 10.0 ± 6.9kg in males and 9.5 ± 6.4kg in females (Figure 3). However, total weight gain had a U-shaped distribution by BMI category in males with median gains of 6.7kg, 5.4kg, 5.0kg, 6.1kg, 6.8kg and 8.6kg across BMI categories. In contrast, females showed a linear increase in total weight gain, with median gains of 5.0kg, 4.6kg, 5.6kg, 6.8kg, 7.3kg and 8.3kg across BMI categories (Figure 4). Male weight-gainers were more likely to have overweight BMI whereas female weight-gainers were more likely to have normal weight BMI at baseline. Given small effects, weight-gainers had higher serum creatinine, eGFR, and triglycerides, and lower HDL than weight-stable participants. A higher percentage of weight-gainers had prescribed anti-hypertensive and anti-diabetic medications compared to weight-stable participants.

Figure 3:

Boxplots Presenting Total and Maximum Weight Change by Weight Trajectory Group

3.5. Weight-Cycling Group

Of 83,261 participants, 50.4% of males and 57.0% of females experienced at least one full weight cycle (Supplemental Table 2). The maximum number of weight cycles across 5 years was 6.0 using the 5% cut-point and 5.0 using the 10% cut-point. Median weight change per cycle was 17.3 (13.4, 23.4) kg in males and 15.5 (11.4, 21.3) kg in females. Using the ≥10% threshold, average weight change per cycle was 29.0 (23.3, 37.5) kg in males and 24.9 (19.6, 33.0) kg in females. The average amount of weight change in weight-cyclers decreased across BMI categories. At the end of the 5-year period, weight cycling yielded weight gain ≥5% of baseline in 26% of males and 30% of females and weight loss ≥5% in 22% of males and 24% of females. Thus, 52% of male and 46% of female weight cyclers experienced no significant change from baseline to final body weight over 5 years.

Average baseline BMI did not differ among groups; 41% of male and 36% of female weight cyclers had BMIs in the obese category at baseline. Weight-cyclers were more likely to have a history of smoking despite being younger than weight-stable participants (males: 50.2 ± 15.2 vs 54.5 ± 13.2 years, p=2.59E-55; females: 47.9 ± 16.0 vs 53.8 ± 14.0 years, p=8.95E-107) and the proportion of weight-cyclers declined across age quartiles. In male weight-cyclers, blood glucose, creatinine, and triglycerides were higher, and HDL-cholesterol was lower than all other groups. In female weight-cyclers, blood glucose and triglycerides were higher than weight-stable and weight-gainers. Females with weight cycling had lower HDL-cholesterol than any other group. Compared to weight-stable and weight-gainers, weight-cyclers had more prescribed dyslipidemia and antidiabetic therapies. There was a small effect (ε² 0.02) for higher cardiometabolic risk score in weight-cyclers compared to other groups (Table 4).

4. Discussion

A notable finding of the present work is that only 8–12% of a cohort representing a mid-South regional comprehensive care medical center were weight-stable during the 5-year observation period, indicating that clinically significant weight fluctuations are highly prevalent in today’s patient population. Interestingly, weight stability was more common in males and the proportion of participants with weight stability increased with age. Males with weight stability were more likely to have overweight BMI while females with weight stability were more likely to have normal weight BMI. Weight-stable participants were more likely to be non-smokers, had lower triglycerides and higher HDL-cholesterol levels, and were less likely to have prescribed medications for dyslipidemia, hypertension, or hyperglycemia. The clinical characteristics of participants with weight stability, even among those with overweight BMI, supports the concept of shifting the prevailing weight management paradigm from being weight-focused to cardiometabolic health-focused.²⁶ It is also plausible that persons with better cardiometabolic health are less vulnerable to future weight gain or cycling as they may engage in healthier diets, physical activities, and other risk reduction behaviors.

The argument for promoting weight maintenance when clinical biomarkers indicate good health, i.e., absent chronic cardiometabolic disease, is strengthened by the potential for adverse health consequences with weight cycling. Weight cycling was the most common trajectory identified, occurring in 50% of males and 57% of females. ASV was significantly higher in weight-cyclers, nearly 2-fold that of weight-losers or gainers and almost 3-fold that of weight-stable participants. However, with the upper quartile of ASV being 1.11 in males and 1.01 in females, ASV represented only half of weight-cyclers in this cohort. Weight cycling was more prevalent among those with a smoking history and younger participants, but the frequency of weight cycling declined with age. Interestingly, the amount of weight change among weight-cyclers decreased with increasing BMI. Regardless of BMI, ~50% of weight-cyclers achieved no overall loss or gain from baseline to final weight over the 5-year period. Nevertheless, weight-cyclers were more likely than weight-stable, weight-gain or weight-loss participants to have prescribed antihypertensive, dyslipidemia, or antidiabetic therapies, indicating greater prevalence of cardiometabolic disease. Indeed, overall cardiometabolic risk score was highest in weight-cyclers. However, the magnitude of these effects was small which may, at least partly, explain why human studies have conflicting results. For example, a 30-month follow up of 153 adults with overweight exposed to caloric restriction-induced weight loss, showed no effects of weight cycling on blood pressure or lipids.²⁷ Nevertheless, several studies show weight cycling is associated with increased risk for coronary heart disease and for cardiovascular and all-cause mortality.^13,28,29 Additionally, meta-analysis showed weight cycling is an independent predictor for new-onset diabetes.⁷ The present findings enhances support for increased cardiometabolic risk particularly considering the 5-year observation period incorporated - the timeframe in which most weight regain occurs.^4,30 Moreover, preclinical evidence suggests weight cycling promotes adipose tissue inflammation and impairs systemic glucose tolerance and insulin sensitivity.^14,17 Thus, from a public health perspective, weight cycling may be considered an additional risk factor for cardiometabolic disease.

As expected, weight-gainers were more common than weight-losers with 23% of males and females being weight-gainers compared to 14% of males and 13% of females being weight-losers. Participants with weight gain were younger, but baseline BMI did not differ across weight trajectory groups. The pattern of weight gain increased linearly across BMI categories in females, ranging from a median of 4.6 to 8.3kg in 5 years. However, the pattern was more U-shaped in males, with median weight gain of 6.7kg in the underweight BMI group, 5.0kg in the overweight BMI group, and 6.1–8.6kg in the obesity BMI groups. Weight-gainers had a small effect for higher triglyceride levels compared to weight-stable group but did not have a higher percentage of participants with prescribed medications for dyslipidemia.

Although the percentage of participants categorized as weight-loser was small, large weight losses ≥10% baseline weight occurred in 30% of males and 42% of females in this group. Like prior evidence, the majority of those with weight loss experienced weight regain at some point during the 5-year period. Indeed, meta-analysis of 20 weight loss studies showed 70–80% of lost weight is regained within 5 years.³⁰ However, other systematic reviews show maintaining 5% or 10% weight loss over time is possible for at least 61% of individuals, particularly when interventions incorporate diet and physical activity strategies.^31,32 It is noteworthy that weight-losers had higher glucose and triglyceride levels and greater percentage of prescribed antihypertensive, dyslipidemia, and antidiabetic therapies than weight-gainers or weight-stable participants, suggesting that cardiometabolic health rather than weight or BMI may be a strong driver for losing weight.

Potentially complicating the ability to precisely identify and characterize weight trajectories, especially weight cycling, is the heterogeneity of definitions used. Most commonly, weight cycling is defined by the absolute amount of weight loss and regain – although even the amounts considered meaningful vary.^12,20,33 In contrast, we used percentage change from each prior weight as a loss or gain of 5–10% is associated with significant changes in blood pressure, glucose, and lipids,^22,34 as well as overall cardiometabolic risk.^35,36 Furthermore, we utilized a comprehensive and iterative methodology to clean weight data for determining trajectories. Moreover, linear interpolation was employed to create a standardized grid of weights which additionally reduces potential error. While ASV is an oft-used measure in weight cycling research, it varies when the number of recorded weights differs, even when weight paths are identical. In the present study, ASV is low compared to other evidence as the number of observations was normalized using the interpolated grid. Here we demonstrate excellent agreement between ASV and weight trajectory categorization using the 5% threshold emphasizing the strength of the categorization in capturing overall weight fluctuations. Another strength of the study is that all available patient data from a large regional healthcare system during a prolonged period (1997–2020) was utilized to yield a cohort of 83,261 participants – a sample size which enables inferences about population level trends. This cohort had similar demographic characteristics to a previously published EHR cohort,³⁷ strengthening the reliability of the findings. A limitation of the study is the inability to determine voluntary versus involuntary weight loss. Beyond the linear interpolation to eliminate frequently recorded weights, we also excluded patients with cancer diagnoses (other than non-melanoma skin cancer) to reduce potential bias from conditions promoting involuntary weight loss. While the effect sizes reported for our findings were of low to moderate in magnitude, interpretation of effect sizes is affected by multiple factors including measurement units, data variability, and study population. Additionally, effect may be modified by variables, some of which can be accounted for by stratifying data as was done for age, sex and BMI. However, the systematic bias inherent in using existing EHR data limits the completeness of laboratory data, constrains access to non-recorded weight fluctuations, and impedes capturing information such as dietary intakes and physical activity which may influence weight trajectories. Thus, generalizability of the findings may be limited to similar healthcare-seeking environments.

Conclusions

The repeated pattern of having at least 5% weight loss and regain (i.e., weight cycling) was highly prevalent within a 5-year observation period in this comprehensive medical center cohort study. Yet, half of weight-cyclers achieved no appreciable change in weight, supporting the construct that sustaining weight loss in the current obesogenic environment is challenging. In comparison to other weight trajectory groups, weight-cyclers had higher cardiometabolic risk profiles, although the magnitude of the effect was small to moderate. Despite having similar baseline high BMI, participants with weight stability had better lipid profiles and lower overall cardiometabolic risk score. These results suggest that maintenance of a stable weight may be preferable for certain individuals. Future work is necessary to determine mechanisms linking weight cycling with cardiometabolic outcomes and to determine efficacious personalized approaches to weight management that target improving cardiometabolic health.

Figure 4e-f:

Total and Maximum Weight Change by Weight Trajectory Group Stratified by BMI Category

Table 1b.

Baseline Demographics by Weight Trajectory Group in Females

	All	Stable	Gain	Loss	Cycle	P-value	Effect Size
Count (n)	50,201 (100%)	3,796 (8%)	11,468 (23%)	6,302 (13%)	28,635 (57%)
Age (years)	49.00 ± 15.78	53.83 ± 13.96a	46.55 ± 14.84b	55.42 ± 15.46c	47.94 ± 15.95d	<0.0001	Є2 = 0.04
Height (cm)	163.59 ± 6.77	163.46 ± 6.56ab	163.94 ± 6.65c	163.20 ± 6.65a	163.56 ± 6.86b	8.94E-11	Є2 < 0.01
Weight (kg)	76.06 ± 19.71	71.48 ± 17.21a	73.38 ± 18.20b	77.96 ± 20.16c	77.32 ± 20.29d	5.92E-136	Є2 = 0.01
BMI (kg/m2)	28.40 ± 7.09	26.75 ± 6.25a	27.29 ± 6.56b	29.27 ± 7.36c	28.87 ± 7.25d	4.95E-167	Є2 = 0.02
BMI Category						9.97E-152	V = 0.07
Underweight	910 (2%)	68 (2%)	258 (2%)	54 (1%)	530 (2%)
Normal Weight	17,954 (36%)	1,786 (47%)	4,753 (41%)	2,000 (32%)	9,415 (33%)
Overweight	14,408 (29%)	986 (26%)	3,294 (29%)	1,884 (30%)	8,244 (29%)
Class I Obesity	8,794 (18%)	547 (14%)	1,764 (15%)	1,207 (19%)	5,276 (18%)
Class II Obesity	4,553 (9%)	273 (7%)	819 (7%)	585 (9%)	2,876 (10%)
Class III Obesity	3,582 (7%)	136 (4%)	580 (5%)	572 (9%)	2,294 (8%)
Race (self-reported)						1.28E-20	V = 0.03
White	42,843 (85%)	3,330 (88%)	9,833 (86%)	5,441 (86%)	24,239 (85%)
Black	5,958 (12%)	318 (8%)	1,287 (11%)	675 (11%)	3,678 (13%)
Other	1,400 (3%)	148 (4%)	348 (3%)	186 (3%)	718 (3%)
Smoking Status						3.62E-73	V = 0.06
Non-Smoker	33,405 (67%)	2,835 (75%)	8,176 (71%)	4,191 (67%)	18,203 (64%)
Ever-Smoker	12,387 (25%)	714 (19%)	2,483 (22%)	1,509 (24%)	7,681 (27%)
Unknown	4,409 (9%)	247 (7%)	809 (7%)	602 (10%)	2,751 (10%)

Data are presented as means with standard deviations for continuous variables and as number with frequency for categorical variables. For continuous variables, Kruskal-Wallis was performed to determine overall group differences with effect size (ε²). This was followed with post-hoc Dunn test employing Benjamini-Hochberg correction to identify between-group differences, denoted using superscript letters. Superscript letters that are not the same between any two weight trajectory groups (e.g., a vs b) denote a significant difference between those two groups at the level of p <0.05. Categorical variables data are displayed as frequency with number and percentage, with p-value as determined by Pearson’s chi-squared and effect size (Cramer’s V). Effect size: ε²: 0.01–0.06 small, 0.06–0.14 moderate, ≥0.14 large effect; Cramer’s V: ≤0.1 weak, 0.3 moderate, ≥0.5 strong association.

Table 2b.

Baseline Demographics by Weight Variability (ASV) Quartiles in Females

	All	Q1	Q2	Q3	Q4	P-value	Effect Size
Count (n)	50,201 (100%)	12,551 (25.0%)	12,550 (25.0%)	12,550 (25.0%)	12,550 (25.0%)
Age (years)	49.00 ± 15.78	50.77 ± 15.36a	49.82 ± 16.18b	48.71 ± 15.98c	46.71 ± 15.30d	1.78E-101	Є2 < 0.01
Height (cm)	163.59 ± 6.77	162.83 ± 6.52a	163.37 ± 6.73b	163.69 ± 6.73c	164.49 ± 6.98d	1.564E-81	Є2 < 0.01
Weight (kg)	76.06 ± 19.71	65.98 ± 13.17a	71.93 ± 15.67b	77.82 ± 17.93c	88.50 ± 23.18d	<0.0001	Є2 = 0.18
BMI (kg/m2)	28.40 ± 7.09	24.90 ± 4.84a	26.96 ± 5.74b	29.06 ± 6.53c	32.70 ± 8.29d	<0.0001	Є2 = 0.17
BMI Category						<0.0001	V = 0.24
Underweight	910 (2%)	382 (3%)	242 (2%)	173 (1%)	113 (1%)
Normal Weight	17,954 (36%)	7,157 (57%)	5,138 (41%)	3,543 (28%)	2,116 (17%)
Overweight	14,408 (29%)	3,322 (26%)	4,027 (32%)	4,023 (32%)	3,036 (24%)
Class I Obesity	8,794 (18%)	1,167 (9%)	1,986 (16%)	2,705 (22%)	2,936 (23%)
Class II Obesity	4,553 (9%)	388 (3%)	760 (6%)	1,285 (10%)	2,120 (17%)
Class III Obesity	3,582 (7%)	135 (1%)	397 (3%)	821 (7%)	2,229 (18%)
Weight Group						<0.0001	V = 0.37
Stable	3,796 (8%)	3,339 (27%)	410 (3%)	47 (0%)	0 (0%)
Gainer	11,468 (23%)	4,770 (38%)	3,833 (31%)	2,168 (17%)	697 (6%)
Loser	6,302 (13%)	2,635 (21%)	2,207 (18%)	1,108 (9%)	352 (3%)
Cycler	28,635 (57%)	1,807 (14%)	6,100 (49%)	9,227 (74%)	11,501 (92%)
Race (self-reported)						1.05E-235	V = 0.11
White	42,843 (85%)	11,177 (89%)	10,939 (87%)	10,670 (85%)	10,057 (80%)
Black	5,958 (12%)	800 (6%)	1,245 (10%)	1,616 (13%)	2,297 (18%)
Other	1,400 (3%)	574 (5%)	366 (3%)	264 (2%)	196 (2%)
Smoking Status						6.2E-151	V = 0.08
Non-Smoker	33,405 (67%)	9,268 (74%)	8,625 (69%)	8,177 (65%)	7,335 (58%)
Ever-Smoker	12,387 (25%)	2,367 (19%)	2,894 (23%)	3,270 (26%)	3,856 (31%)
Unknown	4,409 (9%)	916 (7%)	1,031 (8%)	1,103 (9%)	1,359 (11%)

ASV: average successive weight variability. Data are presented as means with standard deviations for continuous variables and as number with frequency for categorical variables. For continuous variables, Kruskal-Wallis was performed to determine overall group differences with effect size (ε²). This was followed with post-hoc Dunn test employing Benjamini-Hochberg correction to identify between-group differences, denoted using superscript letters. Superscript letters that are not the same between any two weight trajectory groups (e.g., a vs b) denote a significant difference between those two groups at the level of p <0.05. Categorical variables data are displayed as frequency with number and percentage, with p-value as determined by Pearson’s chi-squared and effect size (Cramer’s V). Effect size: ε²: 0.01–0.06 small, 0.06–0.14 moderate, ≥0.14 large effect; Cramer’s V: ≤0.1 weak, 0.3 moderate, ≥0.5 strong association.

Table 3b.

Clinical Biomarkers and Prescribed Cardiometabolic Medications by Weight Variability (ASV) Quartiles in Females

	All	Q1	Q2	Q3	Q4	P-value	Effect Size
Count (n)	50,201 (100%)	12,551 (25.0%)	12,550 (25.0%)	12,550 (25.0%)	12,550 (25.0%)
Medications
Antihypertensive Therapy	19,257 (38%)	3,109 (25%)	4,356 (35%)	5,279 (42%)	6,513 (52%)	<0.0001	V = 0.26
Dyslipidemia Therapy	5,829 (12%)	953 (8%)	1,346 (11%)	1,558 (12%)	1,972 (16%)	7.2903E-91	V = 0.09
Antidiabetic Therapy	11,159 (22%)	1,166 (9%)	2,102 (17%)	3,064 (24%)	4,827 (38%)	<0.0001	V = 0.20
Clinical Biomarkers
Glucose (mg/dL)	100.20 ± 37.66	94.73 ± 24.27a	97.46 ± 30.75b	100.98 ± 37.47c	106.79 ± 50.26d	1.13E-82	Є2 < 0.01
Createnine (mg/dL)	0.86 ± 0.60	0.80 ± 0.24a	0.82 ± 0.38b	0.85 ± 0.49c	0.96 ± 0.96d	1.72E-27	Є2 < 0.01
EGFR (mL/min)	86.15 ± 24.03	86.10 ± 21.10a	86.36 ± 22.66a	86.39 ± 24.15a	85.77 ± 27.35a	0.23	Є2 < 0.01
Total Cholesterol	193.99 ± 44.49	195.57 ± 41.14a	195.28 ± 42.57a	193.61 ± 45.61b	191.70 ± 47.89c	9.092E-16	Є2 < 0.01
LDL-Cholesterol	108.39 ± 38.57	108.73 ± 35.39a	109.55 ± 37.80a	108.56 ± 39.40a	106.81 ± 41.11b	1.89E-04	Є2 < 0.01
HDL-Cholesterol	59.24 ± 19.06	65.19 ± 19.74a	60.83 ± 18.55b	57.57 ± 18.27c	53.95 ± 17.93d	<0.0001	Є2 = 0.04
Triglycerides	126.33 ± 88.25	107.72 ± 70.26a	121.57 ± 84.98b	129.60 ± 86.34c	144.47 ± 102.72d	2.07E-222	Є2 = 0.02
Cardiometabolic Risk Score	2.15 ± 1.47	1.32 ± 1.27a	1.89 ± 1.36b	2.37 ± 1.37c	2.90 ± 1.37d	<0.0001	Є2 = 0.07

ASV: average successive weight variability, EGFR: estimated golmerular filtration rate, LDL: low-density lipoprotein, HDL: high-density lipoprotein. Data are presented as means with standard deviations for continuous variables and as number with frequency for categorical variables. For continuous variables, Kruskal-Wallis was performed to determine overall group differences with effect size (ε²). This was followed with post-hoc Dunn test employing Benjamini-Hochberg correction to identify between-group differences, denoted using superscript letters. Superscript letters that are not the same between any two weight trajectory groups (e.g., a vs b) denote a significant difference between those two groups at the level of p <0.05. Categorical variables data are displayed as frequency with number and percentage, with p-value as determined by Pearson’s chi-squared and effect size (Cramer’s V). Effect size: ε²: 0.01–0.06 small, 0.06–0.14 moderate, ≥0.14 large effect; Cramer’s V: ≤0.1 weak, 0.3 moderate, ≥0.5 strong association. Full list of medications available in Supplemental Table 1.

Table 4b.

Clinical Biomarkers and Prescribed Cardiometabolic Medications by Weight Trajectory Group in Females

	All	Stable	Gain	Loss	Cycle	P-value	Effect Size
Count (n)	50,201 (100%)	3,796 (8%)	11,468 (23%)	6,302 (13%)	28,635 (57%)
Medications
Antihypertensive Therapy	19,257 (38%)	1,035 (27%)	3,580 (31%)	2,527 (40%)	12,115 (42%)	8.33E-139	V = 0.11
Dyslipidemia Therapy	5,829 (12%)	317 (8%)	996 (9%)	846 (13%)	3,670 (13%)	3.56E-42	V = 0.06
Antidiabetic Therapy	11,159 (22%)	409 (11%)	1,754 (15%)	1,549 (25%)	7,447 (26%)	6.98E-187	V = 0.13
Clinical Biomarkers
Glucose (mg/dL)	100.20 ± 37.66	96.05 ± 27.47a	97.11 ± 31.97a	102.99 ± 36.31b	101.27 ± 40.75c	5.86E-48	Є2 < 0.01
Createnine (mg/dL)	0.86 ± 0.60	0.81 ± 0.20a	0.83 ± 0.52b	0.85 ± 0.39c	0.88 ± 0.69a	1.70E-18	Є2 < 0.01
EGFR (mL/min)	86.15 ± 24.03	83.89 ± 20.66a	88.40 ± 22.71b	82.39 ± 22.64c	86.38 ± 25.05d	6.11E-66	Є2 < 0.01
Total Cholesterol	193.99 ± 44.49	196.54 ± 40.74a	194.42 ± 42.31bc	194.83 ± 43.35b	193.32 ± 45.98c	0.00000552	Є2 < 0.01
LDL-Cholesterol	108.39 ± 38.57	109.76 ± 36.05ab	109.75 ± 37.06a	108.07 ± 38.86b	107.75 ± 39.37b	1.74E-03	Є2 < 0.01
HDL-Cholesterol	59.24 ± 19.06	64.34 ± 20.30a	60.61 ± 18.55b	59.64 ± 19.41c	57.97 ± 18.86d	1.02E-73	Є2 < 0.01
Triglycerides	126.33 ± 88.25	112.33 ± 74.16a	118.52 ± 80.60b	128.29 ± 84.65c	130.73 ± 92.99c	6.68E-47	Є2 < 0.01
Cardiometabolic Risk Score	2.15 ± 1.47	1.51 ± 1.36a	1.80 ± 1.38b	2.21 ± 1.47c	2.35 ± 1.46d	2.4058E-191	Є2 = 0.02

EGFR: estimated golmerular filtration rate, LDL: low-density lipoprotein, HDL: high-density lipoprotein. Data are presented as means with standard deviations for continuous variables and as number with frequency for categorical variables. For continuous variables, Kruskal-Wallis was performed to determine overall group differences with effect size (ε²). This was followed with post-hoc Dunn test employing Benjamini-Hochberg correction to identify between-group differences, denoted using superscript letters. Superscript letters that are not the same between any two weight trajectory groups (e.g., a vs b) denote a significant difference between those two groups at the level of p <0.05. Categorical variables data are displayed as frequency with number and percentage, with p-value as determined by Pearson’s chi-squared and effect size (Cramer’s V). Effect size: ε²: 0.01–0.06 small, 0.06–0.14 moderate, ≥0.14 large effect; Cramer’s V: ≤0.1 weak, 0.3 moderate, ≥0.5 strong association. Full list of medications available in Supplemental Table 1.

Study Importance Questions.

• What is already known about the subject?

  • Prior studies have shown that weight cycling (sometimes called ‘weight fluctuation’ or ‘yo-yo dieting’) is a common phenomenon, although the definitions, measurement units, sample sizes, and length of follow-up employed vary.
  • Weight cycling may carry greater risk for cardiometabolic disease; however, findings from human studies are conflicting.
• What are the new findings in your manuscript?

  • Using de-identified EHR data from a cohort of 83,261 patients, we identified 54% experienced weight cycling, but no net change in weight occurred over the 5-year observation period.
  • Weight-cyclers had worse blood glucose, HDL, and triglycerides levels, and significantly greater overall cardiometabolic risk score, despite being younger and having no difference in baseline BMI as weight-gainers, weight-losers, or weight-stable participants.
• How might your results change the direction of the research of the focus of clinical practice?

  • Weight cycling should be considered a potential predictor of cardiometabolic risk and deserves consideration in clinical weight management practices.
  • This study demonstrates the need for further investigation to determine the relationship and mediating factors between weight cycling and cardiometabolic disease outcomes in large representative cohorts.

Data Availability

Data described in the manuscript and analytic code will be made available upon reasonable request and completion of a data-use agreement.