Impact of Weight Loss and Regain on Quality of Life: Mirror Image or Differential Effect?

Authors


Neuropsychiatric Research Institute, 700 First Avenue South, Fargo, ND 58107. E-mail: scott.engel@ndsu.nodak.edu

Abstract

Objective: To compare the impact of weight regain and weight loss on health-related quality of life.

Research Methods and Procedures: Subjects were 122 (106 women, 16 men) overweight and obese participants in a weight reduction program (phentermine-fenfluramine and dietary counseling) who had initially lost at least 5% of their total body weight and then regained at least 5% of their weight during the follow-up period. Follow-up periods ranged from 10 to 41 months (mean, 28 months). Participants completed the Impact of Weight on Quality of Life-Lite, an obesity-specific health-related quality of life (HRQOL) measure, at 3-month intervals.

Results: Mean BMI at baseline was 40.9 ± 6.6 kg/m2 (range, 29.2 to 63.7 kg/m2). Average weight loss from entry was 18.8 ± 6.7% (range, 6.0% to 43.7%), and average regain was 10.1 ±4.4% of baseline weight (range, 5.0% to 30.6%). The effects of weight regain on HRQOL mirrored the effects of weight loss—rates of HRQOL change were similar in magnitude but different in direction for comparable weight loss and regain. Those with more severe initial impairments in HRQOL experienced greater improvements in HRQOL during weight loss as well as greater deterioration during weight regain than those with less severe impairments.

Discussion: Weight loss and regain produced mirror image changes in HRQOL. The initial severity of HRQOL impairment had a greater impact on the magnitude of HRQOL change than the direction of weight change. Findings underscore the importance of maintaining weight loss for the purposes of retaining obesity-specific HRQOL benefits.

Introduction

The prevalence of obesity is increasing, with an estimated prevalence of 250 million people worldwide, which is equivalent to 7% of the adult population (1). The most recent survey (1999 to 2000) from the National Health Examination Survey reports that the age-adjusted prevalence of overweight adults in the United States is 64.5% and of obese adults is 30.5% (2). Lysne et al. (3) report that obesity rates in the United States increased from 12% to 18% in just a 7-year span in the 1990s. This is of great concern given that obesity is also associated with a number of negative health outcomes, such as increased risk for type 2 diabetes, hypertension, gallbladder disease, and osteoarthritis (4).

There have been numerous reports in the literature, from both population studies and weight loss studies, suggesting that obesity is associated with impaired health-related quality of life (HRQOL)1 (5, 6, 7, 8). Furthermore, weight loss, in those who are overweight or obese, has been associated with improvements in HRQOL (9, 10, 11, 12, 13, 14).

However, there is a marked paucity of research on the effects of weight gain on HRQOL. To our knowledge, only one study addresses this issue. In a population study of U.S. nurses over a 4-year period, weight gain, regardless of starting BMI, was associated with impairments in HRQOL (14).

To the best of our knowledge, no research to date has studied the impact of weight regain after weight loss on HRQOL. Because it is unknown how weight regain affects HRQOL, this study examines this relationship. Specifically, we address the following questions.

  • 1. Does the loss of weight result in improved HRQOL to the same extent that the regaining of weight results in decreased HRQOL?
  • 2. Is the relationship between HRQOL and weight change (both weight loss and weight regain) a linear or curvilinear path?
  • 3. Does initial severity of baseline HRQOL impact the manner in which HRQOL changes with respect to weight loss and regain?

Portions of this paper were presented at the 12th European Congress on Obesity, Helsinki, Finland, May 2003.

Research Methods and Procedures

Participants

Participants in this report were selected from 199 overweight and obese individuals (170 women, 29 men) in an obesity treatment program conducted at Hennepin County Medical Center, Minneapolis, Minnesota, between October 1993 and December 1997. All individuals in this program were at least 130% of their ideal body weight at entry. Other medical and psychological exclusion criteria are detailed elsewhere (9, 10). Data in this report are based on the 122 individuals (106 women, 16 men) in that program who initially lost at least 5% and subsequently regained at least 5% of their baseline weight. The 122 included participants did not differ significantly from the remaining 77 who were excluded as to gender (86.9% vs. 83.1% women, Fishers exact p = 0.537), age (45.6 ± 9.1 vs. 44.0 ± 9.9 years, t = −1.17, df = 197, p = 0.242), baseline BMI (41.0 ± 6.7 vs. 40.0 ± 8.3 kg/m2, t = −0.82, df = 197, p = 0.388), or baseline Impact of Weight on Quality of Life-Lite (IWQOL-Lite; described below) total score (74.0 ± 15.8 vs. 77.1 ± 15.6, t = 1.37, df = 195, p = 0.173).

Assessment Instruments

Participants completed a self-report measure of obesity-specific quality of life, the IWQOL (15, 16). The original IWQOL is a 74-item version. Data presented in this report were scored using the newer, more psychometrically sound 31-item version (9, 10), the IWQOL-Lite. IWQOL-Lite scores range from 0 (worst possible quality of life) to 100 (best possible quality of life). There are five subscales on the IWQOL-Lite (Physical Function, Self-Esteem, Sexual Life, Public Distress, Work) and a total score. In previous studies, the IWQOL-Lite demonstrated excellent psychometric properties: internal consistency ranged from 0.90 to 0.94 for scales and was 0.96 for total score (9); test-retest reliability ranged from 0.81 to 0.88 for scales and was 0.94 for total score (17); the scale structure was confirmed by factor analysis (9); and there was good support for construct validity in that scales correlated well with BMI (9, 10, 17, 18), weight loss (10), treatment-seeking status (18), and appropriate collateral measures (9, 17).

Procedures

Participants completed a thorough medical history and physical examination conducted by a study internist at the beginning of the study, a psychological assessment and structured diagnostic interview conducted by a study psychologist, a nutritional assessment conducted by a registered dietitian, and an extensive laboratory evaluation. For all 122 participants in this study, the treatment program consisted of regular dietary counseling and individualized diets, exercise recommendations, and open-label medication. Medication was d,l-fenfluramine HCl (20 mg orally three times a day up to a maximum of 120 mg/d) combined with phentermine HCl (30 mg orally daily). Participants completed the IWQOL and were weighed at baseline and 3-month intervals thereafter for a maximum of nearly 3.5 years.

Statistical Analysis

Data were accumulated for all participant visits at which both weight and a valid IWQOL-Lite total score were obtained. The IWQOL-Lite score at each visit was expressed as a change score from the previous visit. Weight change from the previous visit was expressed as a percentage of baseline weight.

A series of random effects linear model (19) analyses were then performed to examine the relationship between change (and direction of change) in weight and change in IWQOL-Lite total score. Analyses were completed on the subscale scores of the IWQOL-Lite, and these results were found to be comparable to those using the total score. All data presented in the present study are from use with the IWQOL-Lite's total score. All analyses were performed using SAS PROC MIXED (20). Conceptual models of change that were tested in these analyses are illustrated in Figure 1.

Figure 1.

Conceptual models of change: model 1 demonstrates a “mirror-image” relationship between HRQOL change and weight loss and gain. Model 2 demonstrates a differential effect across weight loss and regain. Models 3 and 4 demonstrate a curvilinear relationship for weight loss and gain, respectively. Model 5 shows HRQOL change for weight loss and regain as a function of initial severity of patients’ impaired HRQOL.

The random regression approach used in the present analyses has several advantages over more traditional analytic approaches to longitudinal data analysis (e.g., repeated measures ANOVA, MANOVA). These include 1) the inclusion of subjects that terminate early or that have missing data without relying on data imputation procedures, 2) the use of both fixed and time-varying covariates, 3) the measurement of subjects at different time intervals, 4) the assessment of individual change estimates for each subject, and 5) the use of serial-dependent error components to account for the covariation of repeated observations.

A prerequisite to testing more complex models (models2–5) would be to demonstrate that there is a significant relationship between weight change and HRQOL. Model 1 evaluated whether there was a significant linear relationship between percent weight change and change in quality of life across the entire range of weight change. To test for mirror-image or differential effects across weight loss and regain, model 2 tested whether the slope differed for weight loss and weight regain. The model included a random effect(s) for subject (intercept) and a fixed effect(s) for weight change (slope) and loss/gain. The next models evaluated whether there was a significant linear or curvilinear relationship separately for weight loss (model 3) or gain (model 4). These were completed separately so that we could test for linearity in one direction of weight change (e.g., loss), but curvilinearity in the other direction of weight change (e.g., gain). The models included a random effect(s) for subject (intercept) and fixed effect(s) for weight change (linear) and squared weight change (curvilinear). The squared weight change model was chosen because of its reflexive property such that weight gain and loss could be modeled as mirror images. In addition, these models are among the simplest forms of curvilinear models and may provide reasonable approximations for more complex nonlinear models (21). The final model (model 5) tested whether the slope is significantly different according to severity of baseline quality of life. Subjects were classified into three baseline severity groups (low, medium, high) based on whether their baseline IWQOL-Lite total score was the following: 1) <1 SD (low: n = 47); 2) 1 to 2 SDs (medium, n = 29); or 3) more than 2 SDs (high, n = 46) of the mean gender-specific score of a normative sample of 615 overweight (BMI, 25 to 30 kg/m2) participants (17, 18). Model 5 contained a random effect(s) for subject (intercept) and fixed effect(s) for weight loss, severity, and weight loss by severity interaction.

Results

Demographics

The current sample consisted of 106 women (86.9%) and 16 men (13.1%). Mean age of participants at entry into the study was 45.6 ± 9.1 years (range, 25 to 65 years). Baseline BMI was 41.0 ± 6.7 kg/m2 (range, 29.2 to 62.6 kg/m2). The mean BMI was 40.7 ± 6.5 kg/m2 (range, 29.2 to 62.6 kg/m2) for women and 42.8 ± 7.7 kg/m2 (range, 33.7 to 60.1 kg/m2) for men. The observation period for subjects ranged from 314 to 1239 days (842 ± 201 days).

Weight Loss and Regain

The average weight loss from baseline to low weight was 18.8 ± 6.7% (range, 6.0% to 43.7%) and did not differ by gender (women, 18.9 ± 6.5%; men, 18.3 ± 8.1%). Lowest weight occurred at 367 ± 114 days after baseline (range, 106 to 749 days). The average weight regain was 10.1 ± 4.4% (range, 5.0% to 30.6%) and did not differ by gender (women, 10.1 ± 4.2%; men, 10.3 ± 6.0%). Highest weight during regain occurred 808 ± 199 days after baseline (range, 314 to 1190 days). For a summary of BMI (and percentage weight change) by group severity, see Table 1.

Table 1.  BMI and percent weight change by baseline severity
 Mild (n = 47)Moderate (n = 29)Severe (n = 46)Overall (n = 122)
Baseline BMI37.1 ± 4.643.1 ± 7.143.4 ± 6.240.9 ± 6.6
Percent change in baseline weight to low weight−19.2 ± 6.8−18.1 ± 6.0−18.9 ± 7.1−18.8 ± 6.7
Low BMI29.9 ± 4.035.4 ± 6.735.2 ± 5.833.2 ± 6.0
Percent change in low weight to high weight10.1 ± 4.610.1 ± 3.810.2 ± 4.710.1 ± 4.4
High BMI33.7 ± 4.839.7 ± 6.839.6 ± 6.237.3 ± 6.5

Quality of Life Scores

Mean IWQOL-Lite total score at baseline was 74.1 ± 15.6 (range, 23 to 100), at low weight was 87.3 ± 11.7 (range, 38 to 100), and at high weight was 82.5 ± 15.3 (range, 26 to 100). Comparison of these scores with repeated measures ANOVA and Bonferroni post hoc comparisons revealed significant differences (F(1121) = 49.98, p < 0.001), with each score different from all others at p < 0.001. For a summary of HRQOL scores by group severity, see Table 2.

Table 2.  IWQOL-Lite scores by baseline severity
 Mild (n = 47)Moderate (n = 29)Severe (n = 46)Overall (n = 122)
Baseline88.3 ± 5.875.9 ± 4.958.6 ± 12.374.1 ± 15.6
Δ Baseline to low weight6.5 ± 7.011.6 ± 8.821.1 ± 13.613.2 ± 12.1
Low weight94.7 ± 4.787.5 ± 8.579.6 ± 13.687.3 ± 8.5
Δ Low weight to high weight−2.4 ± 6.1−5.2 ± 11.5−7.0 ± 11.2−4.8 ± 9.8
High weight92.3 ± 7.182.3 ± 13.472.6 ± 16.282.5 ± 15.3

Model Testing

Model 1 was used to test whether there was a significant linear relationship between change in weight and change in quality of life. As expected, there was a strong linear relationship between these variables (estimate = −0.708, SE = 0.045, t(854) = −15.59, p < 0.0001). The results from model 2 indicated that the slopes did not differ for loss vs. regain (estimate = 0.008, SE = 0.173, t(853) = 0.05, p = 0.962).

Model 3 was used to test whether there was a significant linear or curvilinear relationship between weight loss (i.e., excluding weight regain) and improvements in quality of life. There was a strong linear relationship (estimate = −0.785, SE = 0.080, t(322) = −9.83, p < 0.0001) but no evidence of a curvilinear relationship (estimate = 0.006, SE = 0.013, t(321) = 0.49, p = 0.626).

Model 4 was used to test whether there was a significant linear or curvilinear relationship between weight regain (i.e., excluding weight loss) and deterioration in quality of life. Again, there was a strong linear relationship (estimate = −0.917, SE = 0.145, t(409) = −6.32, p < 0.0001) but no evidence of a curvilinear relationship (estimate = −0.049, SE = 0.047, t(408) = −1.03, p = 0.302).

Model 5 was used to test whether slopes differed as a function of baseline severity, as measured by the IWQOL-Lite. The slopes of both the medium severity (estimate = 0.615, SE = 0.116, t (852) = 5.29, p < 0.0001) and high severity (estimate = 0.797, SE = 0.100, t(852) = 7.99, p < 0.0001) groups were significantly greater than the slope of the low severity group.

Discussion

To briefly summarize our findings, our participants both lost and regained a considerable amount of weight. They also experienced substantial amounts of HRQOL change associated with these weight changes. Weight loss was associated with improvement in HRQOL, and weight regain was associated with deteriorations in HRQOL. Also, these changes in HRQOL occurred in a linear fashion across the entire weight change continuum from weight loss to weight regain (see Figure 1, model 1). Importantly, weight loss and weight regain were similar in degree, but opposite in the direction, of HRQOL change. In other words, participants’ HRQOL improved to the same extent for each unit of weight lost as it deteriorated for each unit of weight regained. Finally, those patients who initially reported the greatest impairment in HRQOL also reported greatest improvement in HRQOL per unit of weight lost and greatest deterioration of HRQOL per unit of weight regained. Conversely, those who reported less HRQOL impairment initially reported less improvement per unit of weight change and also reported less deterioration of HRQOL per unit of weight regained.

Other non-obesity-related research has addressed whether improvement or deterioration of a disease state has mirror-image or differential effects on a person's HRQOL. Some research suggests that a smaller amount of change may be more meaningful when a patient is improving compared with worsening (thus, suggesting differential rather than mirror-image effects). Cella et al. (22), for example, found that cancer patients who reported global worsening over a 2- to 3-month period also reported larger HRQOL change scores than those reporting comparable global improvements. This suggests that those who believe they are getting better require less improvement in HRQOL to categorize themselves as “improved” than those who feel worse need to categorize themselves as “worse.” Cella et al. are not the only researchers to find differences in clinically meaningful HRQOL changes between deterioration and improvement. Ware et al. (23) and Juniper et al. (24) reported similar changes for improvement and deterioration in other groups of patients. We are left to speculate about why Cella et al. found differential effects, whereas the present study revealed mirror-image effects. This could, in part, be because of the vastly different disease states being studied [cancer (22) vs. obesity (the present study) vs. asthma (24)].

This study did find that, across both weight loss and weight regain, patients whose initial HRQOL was most impaired benefited most from weight loss (compared with those who had less severe initial HRQOL), but also lost those benefits as quickly as they gained them during weight regain. This could occur as a clinically meaningful phenomenon or could occur because of simple regression toward the mean (25). If it is not simply caused by regression toward the mean, then those who are most impaired initially will benefit more (than those less initially impaired) for each unit of weight lost. However, their HRQOL improvements will also deteriorate more quickly with each unit of weight regained.

This study offers two primary strengths that are worth noting. First, little or no research has investigated the impact of weight regain on HRQOL. This oversight in past research is particularly significant given the fact that weight loss interventions are remarkably ineffective at helping people to lose weight and maintain their losses (26). Despite some evidence that an individual can lose weight and maintain weight loss (27), it has long been known that most people who lose weight are likely to regain the weight they lost (28). Unfortunately, the high likelihood of weight regain is still a reality today. In this study, we had a large number of participants who both lost and regained a considerable amount of weight. This sample is unique in this respect given that there is very little research on HRQOL in obese samples that have followed participants for a period sufficient to observe such weight losses and regains.

A second strength of the current paper is related to the analyses that were used. Through the use of modern statistical analyses such as hierarchical linear modeling, one can investigate the impact of both loss and regain within a group of individuals rather than comparing across groups of “weight gainers” and “weight losers” (14). Combining the findings and methodologies/analyses of Fine et al. with those of this study, one can reasonably conclude that across groups or within individuals who have changing weight, improved HRQOL is associated with weight loss and decreased HRQOL is associated with weight gain.

A limitation of this study relates to the recruitment process. Participants who were recruited for this study completed a careful screening procedure so that participants who were physically and/or psychologically unhealthy could be excluded. This resulted in a sample that was particularly homogeneous and healthy. This may limit the extent to which one can generalize to other populations of patients. Related to this, the conclusions from this study may not generalize to any obese person who is gaining or losing weight, but rather, may be more appropriately applied to those enrolled in a weight-loss treatment program.

Another limitation that deserves mention is that of the difference or possible difference between weight gain and weight regain. As previously mentioned, there has been only a small amount of research that has investigated the impact of weight gain on HRQOL. Virtually no research has specifically investigated weight regain as opposed to weight gain. The extent to which weight gain and weight regain differ on their impact of HRQOL is unknown at this time. The reader should take caution in generalizing the present findings of weight regain to those of weight gain.

Another limitation could be the fact that the data collection for this study ceased when the Food and Drug Administration announced that research suggested that fenfluramine combined with phentermine was linked with heart valve problems. Thus, a number of subjects may not have been followed for a sufficient window of time to collect long-term HRQOL and weight regain information.

Finally, related to the above mentioned limitation, it is possible that as patients remained on their medication for longer periods of time, heart valve problems may have occurred, thus impacting the patients’ overall HRQOL scores. To investigate this potential limitation, we reviewed the charts of the 122 patients involved in this study. Of these patients, 116 had at least one echocardiogram.

These echocardiograms represent all the available studies done at Hennepin County Medical Center at any time and were performed at a variety of times relative to treatment with fenfluramine and phentermine. A few of them were even performed before the subject was exposed to the drugs. Of these 116 patients, 54 of them had at least one echocardiogram that met Food and Drug Administration case definition of cardiac valvulopothy.

Cardiac valvulopothy has been defined as documented aortic regurgitation of mild or greater severity and/or mitral regurgitation of moderate or greater severity after exposure to these drugs (29). However, when we compared the last available IWQOL-Lite score for those who had a positive echocardiogram (i.e., they met Food and Drug Administration case definition of valve disease on at least one echocardiogram) with those who had all negative echocardiograms, we found no significant differences between the groups (t(114) = 0.33, p = 0.75). Based on these data, it does not seem that the cardiovascular sequelae associated with fenfluramine/phentermine explain the changes seen in the HRQOL scores of participants.

There continues to be a lack of research on weight regain. Whereas this study investigates the impact of weight regain on obesity-specific HRQOL, it is the only study to date to do so. Given the relatively high likelihood of regain for those who have lost weight, further research is needed to investigate the impact of weight regain. Furthermore, the impact of weight regain compared with weight gain is one of interest. Does weight gain impact individuals’ HRQOL in a similar manner that weight regain does? Unanswered questions such as this one are much more abundant than answered questions in this area of the literature. We encourage investigators to continue researching the related areas of weight gain and weight regain and the impact that each has on HRQOL change.

The current findings offer strong empirical support of the importance of weight loss maintenance. Past research has suggested that the deterioration associated with worsening conditions may not be as dramatic as the HRQOL benefits seen in improving conditions (22). However, our data suggest that this is not so for obesity-specific HRQOL and weight loss or weight regain. In fact, our data suggest a linear relationship for HRQOL change across both weight loss and regain. Therefore, patients who have lost weight do not seem to continue to reap the benefits of their successful weight loss once they begin to regain weight.

Acknowledgment

Financial support for this project was provided by Bristol-Myers Squibb, Princeton, NJ.

Footnotes

  • 1

    Nonstandard abbreviations: HRQOL, health-related quality of life; IWQOL, Impact of Weight on Quality of Life.

Ancillary