Will I Get Fat? 22-Year Weight Trajectories of Individuals with Eating Disorders

Abstract

Objective

For some, fat phobia or fear of uncontrollable weight gain is diagnostic of eating disorders, often inhibiting treatment engagement and predicting symptom relapse. Prior research has reported weight changes at infrequent or long intervals, but rate, shape and magnitude of long-term changes remain unknown. Our study examined 22-year longitudinal trajectories of body mass index (BMI) in women with anorexia nervosa (AN) and bulimia nervosa (BN).

Method

Participants were followed over ten years (N=225) and at 22-year follow-up (N=175). Using latent growth curves, we examined: (1) shape and rate of intra-individual BMI change over ten years; (2) predictors of BMI change over ten years, (3) 22-year BMI outcomes; and (4) BMI changes over ten years as predictors of 22-year BMI.

Results

The best-fitting model captured overall intra-individual rates of BMI change in three intervals, showing moderate rate of BMI increase from intake to year two, modest increase from year two to five, and plateau from year five to ten. At 22 years, 14% were underweight, 69% were normal weight, and only 17% were overweight or obese. Greater increases from intake to year two predicted higher BMI at 22 years (β=0.43, p<0.01) and were predicted by intake diagnosis of AN-restricting (β=0.31, p<0.01) or AN-binge eating/purging (β=0.29, p<0.01).

Discussion

BMI increased most rapidly during earlier years of the study for those with lower weight at study intake (i.e., AN) and plateaued over time, settling in the normal range for most. Psychoeducation about expected BMI trajectory may challenge patients’ long-term fat phobic predictions.

For many individuals with eating disorders (EDs), an underlying cognitive mechanism of behavioral symptoms is a fear of becoming fat (i.e., fat phobia) or a fear of weight gain. As life-threatening conditions that are associated with psychiatric comorbidity (1) and often a protracted course (2), EDs can be difficult to treat. Although not all individuals with EDs endorse body image concerns (3, 4), fear of weight gain is associated with a greater likelihood of relapse (5) and premature treatment dropout (6, 7).

Fat phobia and fear of weight gain may influence patients’ perspectives of treatment consequences in both the short-term (e.g., fears of gaining small amounts of weight) and long-term (e.g., fears of precipitous weight gain that will cause them to be “fat” forever) (8). If they stop purging or break a dietary rule, patients may fear that uncontrollable weight gain will ensue. In her memoir of full recovery (9), Jenni Schaefer described her inner struggle with letting go of her eating disorder–“What if being recovered just means that I’m going to gain weight and be fat and miserable?” (9). The pervasive question “will I get fat?” can be a roadblock to behavior change in treatment. Yet, limited research has examined weight trajectories of individuals with EDs over the long-term.

While weight gain is expected with low-weight EDs such as anorexia nervosa (AN) (10–12), data suggest that weight is typically stable in bulimia nervosa (BN) even up to 12-year follow-up (13). In spite of these findings, patients may be fearful of even small weight increases, believing they are harbingers of exponential weight gain over time. Previous long-term follow-up studies of EDs have reported weight outcomes at infrequent or long intervals (10–16). Thus, no published data exist to corroborate or refute patient fears. We can address this knowledge gap by examining weight changes over the long-term and identifying predictors of the rate, direction and magnitude of that course.

Longitudinal trajectories of BMI using latent growth curves (LGC) can elucidate nuances of these intra-individual changes. In a sample of women with AN and BN who were followed intensively for ten years and then re-interviewed at 22 years after intake, we aimed to: (1) plot the rate and shape of intra-individual change in BMI over ten years; (2) explain variation in the rate of change in BMI over ten years; (3) explore BMI outcomes at 22 years; and (4) predict variation in BMI at 22 years.

Method

Participants and procedures

Females (N=250) seeking ED treatment from Boston-area outpatient services were recruited to participate in the Massachusetts General Hospital Longitudinal Study of Anorexia Nervosa and Bulimia Nervosa between 1987 and 1991 (n=4 dropped out after the intake interview and before the first follow-up interview). Criteria for participation included age ≥12 years, residence within 200 miles of Boston, English speaking, no evidence of terminal illness or organic brain syndrome, and DSM-III-R AN or BN (assessed via the Schedule for Affective Disorders and Schizophrenia-Lifetime Version (17), modified to include DSM-III-R diagnostic criteria for AN and BN). Participants were interviewed at 6–12 month intervals through 2000 (“Wave 1”). Between 2011 and 2013 (“Wave 2”), participants were re-contacted 25 years after the start of the study for a one-time follow-up interview, with a mean follow-up length of 22.6 years (SD=1.0 years; range=19.9 to 24.7 years; see Eddy et al. (18)). For our analysis, we excluded participants who were deceased (n=18) or who did not have height and weight data for at least one year (n=3). Thus, the final sample included 225 participants for whom follow-up data were collected for a median of 9 years (range=1–12 years) in Wave 1. Of these 225 participants, 175 (77%) participated in Wave 2. Participants did not receive treatment as part of the study. The study was approved by the Partners Human Research Committee and all participants provided informed consent.

Participant characteristics

Diagnosis at study intake was based on retrospective application of DSM-IV criteria, with 42 AN-restricting type (AN-R) (19%), 77 AN-binge eating/purging type (AN-BP) (34%), and 106 BN (47%). For the AN weight criterion, we used the DSM-IV recommended cut-off of < 85% expected body weight. See Figure 1 for a flow diagram of participant diagnoses and sample selection at each Wave. The average age of ED onset was 17.5 years (range=8–44 years), with a mean illness duration prior to study intake of 6.6 years range=0–25). Participants were primarily middle to upper middle class and Caucasian (19). Mean age at intake (N=225) was 24.0 (SD=6.6) years and mean age at 22-year follow-up (N=175) was 46.3 (SD=6.6) years.

Figure 1.

Flow Chart of Study Participation at Wave 1 and Wave 2 by Intake Diagnosis Note. BN=bulimia nervosa; AN-R=anorexia nervosa, restricting type; AN-BP=anorexia nervosa, binge-eating/purging type.

Body mass index (BMI)

Height and weight were measured by research staff at intake and BMI (weight [kg]/height² [m]) was calculated for each participant and categorized as underweight (<18.5 kg/m²), normal weight (18.5–24.9 kg/m²), overweight (25.0–29.9 kg/m²), or obese (>30.0 kg/m²). For participants who had missing intake measured heights (n=6) and weights (n=2), we calculated BMI with self-reported weights and heights. Correlations were significant between intake-measured and intake self-reported height (r=1.00, p<0.001) and weight (r=0.98, p<0.001), supporting the validity of self-reported measurements in this sample. We used self-reported follow-up weights at each year post-intake during Wave 1 and once at Wave 2. There was no significant change in height from baseline to 22-year self-reported height (t(175)=−0.05, p=0.96).

Covariates

Covariates in the LGC analyses (Aims 2 and 4) included intake diagnosis, age at intake, and illness duration. At each follow-up assessment during Wave 1, participants completed the Longitudinal Interval Follow-Up Evaluation Eating Disorders Version (LIFE-EAT II) (20) with a trained research interviewer. From this assessment, we included recovery status, pregnancy, and hospitalization over Wave 1 as time-varying covariates. The LIFE-EAT II yielded AN and BN psychiatric status ratings assigned by the interviewer (ordinal, symptom-oriented scores based on Research Diagnostic Criteria ratings (see Eddy et al. (18)). Recovery was defined as a score of two or less (indicating minimal or absent symptoms) for at least one year (18) at each year post-intake (no vs. yes). We categorized participants’ pregnancy status and hospitalization status at each year post-intake as yes/no in the past year based on self-report.

Statistical analysis

We estimated three LGC models using structural equation modeling (21). In the first LGC model, we utilized BMI at each year of follow-up to examine the shape and rate of intra-individual change of participants’ BMI during Wave 1 (Aim 1). To obtain the best fitting model, we first tested whether a linear model would account for the change in BMI over time during Wave I. A linear model of change was first examined followed by a non-linear (quadratic) model of change. Difference in fit between the linear and non-linear models was evaluated using a χ² difference test. Based on visual inspection of the observed data, we expected that the non-linear model would provide a better fit to the data relative to the linear model, which we assessed using a chi-square difference test. With a well-fitting non-linear model of change in BMI over time, we further examined the non-linear pattern of change by breaking down the non-linear curve into three linear pieces based on visual inspection of the data. We then conducted a second LGC model, examining how participants’ diagnosis at intake (AN-R, AN-BP, or BN), age at intake, and duration of eating disorder prior to intake predicted BMI scores at intake and change in BMI scores over Wave I. To explore weight outcomes at 22 years, we used a two-factor ANOVA to compare BMI at 22 years among individuals who were recovered or not recovered with Wave 2 diagnoses of AN-R, AN-BP, or BN (Aim 3). Then, in the third LGC model, we examined intake BMI and rates of change in BMI over Wave 1 as predictors of BMI at year 22 (Wave 2) while controlling for participants’ intake diagnosis, age at intake, illness duration (prior to intake) as well as recovery, pregnancy, and hospitalization at each year post-intake (Aim 4).

We used Mplus 7.0 (22) to conduct the LGC analyses. BMI during Wave 1 had modest to moderate positive kurtosis (range: 1.08–11.90) and skewness (range: 0.73–2.40). Thus, we used the Yuan and Bentler (23) scaled likelihood ratio chi-square (YBχ²) to account for deviations from normality in BMI. We also used the comparative fit index (CFI) and root mean square error of approximation (RMSEA) to assess model fit. A good fit is reflected in a non-statistically-significant YBχ², a CFI above 0.95, and a RMSEA less than 0.05 (21).

Missing Data

There were missing data for BMI, recovery, pregnancy, and hospitalization at each follow-up assessment over Wave 1. Missing data ranged from 0% to 8.4% in years 1 to 7, from 18.2% to 20.9% in year 8, from 36.9% to 43.6% in year 9, and from 70.2% to 70.7% in year 10. At Wave II, there was only 22.2% missing data for BMI. Little’s (1988) missing completely at random (MCAR) test was not statistically significant, χ² (147)=161.49, p=0.20, suggesting that the pattern of missing data could be considered MCAR. However, in line with recommendations by Enders (2010), we more carefully assessed whether the missing data were MCAR or missing at random (MAR), by (1) identifying correlates of missingness by comparing participants with missing data to participants with complete data in relation to illness duration and age at intake and (2) examining frequency of missingness at each follow-up as a function of intake ED diagnosis.

Five patterns emerged that suggested the data was MAR. First, participants with missing data on BMI, pregnancy, and hospitalization variables at years 5, 6, 7, and 8 had a shorter duration of eating disorder prior to intake compared to participants with complete data. Second, participants with missing data on BMI, recovery, pregnancy, and hospitalization in years 8 and 9 had a lower BMI at intake compared to participants with complete data. Third, participants with missing data on BMI, hospitalization, and recovery at year 25 had higher BMI at intake as well as higher BMI in years 1, 2, 3, 4, 5, 6, 7, and 8 compared to participants with complete data. Fourth, participants’ diagnosed with AN-R at intake had more missing data on BMI, recovery, pregnancy, and hospitalization in years 8, 9, and 10 compared to participants who were diagnosed with BN or AN-BP at intake. Fifth, participants diagnosed with BN or AN-BP at intake had more missing data on BMI, recovery, and hospitalization at year 25 compared to participants who were diagnosed with AN-R at intake. To handle missing data, we used the full information maximum likelihood estimation method in Mplus. This approach provides less biased parameter estimates compared to listwise deletion (24). Thus, all analyses included the total sample (N=225).

Results

Aim 1: The shape and rate of intra-individual change in BMI over ten years of follow-up

The observed mean trajectory of BMI over Wave 1 (Figure 2 solid line) exhibited a non-linear, increasing pattern of change. As such, we examined if a non-linear (quadratic) pattern of change would provide a good fit to the data. In the LGC model, the adjacent error variance of each observed BMI score was correlated. Initially, the quadratic model provided a close fit to the data, YBχ²(48)=78.82, p=0.003, CFI=0.98, and RMSEA=0.05. There were two residual correlations: (1) between the error variance of BMI at year five and the error variance of BMI at year ten and (2) between the error variance of BMI at year two and the error variance of BMI at year eight. The quadratic model provided an excellent fit to the data after including both residual correlations, YBχ²(46)=52.42, p=0.24, CFI=1.00, and RMSEA=0.03. Including both residual correlations in the model improved model fit, Δχ²(2)=107.63, p<0.01. Figure 2 shows the estimated non-linear trajectory of BMI over Wave 1 (dotted line).

Figure 2.

Shape and Rate of Intra-Individual Change in BMI over Wave 1

Note. BMI=body mass index. The line for the observed trajectory is based on participants without missing data whereas the lines for the non-linear and piecewise trajectories are based on participant with and without missing data.

Based on visual inspection, BMI appeared to change in a piecewise fashion. To further understand the non-linear pattern of change in BMI over Wave 1, we used a piecewise LGC model (25) finding three trajectories of change—linear change from intake to year two, from year two to year five, and from year five to year ten (Figure 2 dashed line). The piecewise model, which included correlations between adjacent error variances and both residual correlations, provided an excellent fit to the data, YBχ²(41)=46.52, p=0.26, CFI=1, and RMSEA=0.02, with a better fit to the data compared to the non-linear (quadratic) model, Δχ²(5)=77.15, p<0.01.

Specifically, across participants, the average intake BMI was 20.21 mg/kg², z=79.67, p<0.01 and there was a moderate increase in average BMI between intake and year two, B=0.50, 95% CI [0.34, 0.66], z=6.17, p<0.01. That is, participants’ BMI increased by 0.50 units per year on average in the first and second years after intake. Between years two and five, average BMI further increased, but at a more modest rate, B=0.18, 95% CI [0.07, 0.29], z=3.13, p<0.01. That is, participants’ BMI increased by 0.18 units per year on average during the third, fourth, and fifth years after intake. Between years five and ten, average BMI remained stable, B=0.01, 95% CI [−0.08, 0.11], z=0.28, p=0.78. Finally, the variance components of intake BMI and each linear piece were statistically significant, suggesting inter-individual differences in intake BMI and in intra-individual change in BMI over Wave 1.

Aim 2: Explaining variation in BMI over ten years of follow-up

An LGC analysis was conducted to examine participants’ intake diagnosis, age at intake, and duration of illness as predictors of intake BMI and change in BMI over Wave 1. To include intake diagnosis in the model, we created two dummy variables. The reference category was BN and we examined the trajectory of participants with AN-R, AN-BP, and BN. In addition, we controlled for three time-varying covariates: recovery (no/yes), pregnancy (no/yes), and hospitalization due to one’s ED (no/yes) at each year post-intake. To enhance the accuracy of parameter estimates, recovery, pregnancy, and hospitalization at each follow-up period were regressed on illness duration and age at intake.

Intake diagnosis predicted intake BMI and change in BMI from intake to year two (Table 1). As expected, simple slopes analyses (26) indicated that AN-R and AN-BP participants had lower average intake BMI compared to BN participants, and that AN-R and AN-BP participants’ BMI increased at a moderate rate from intake to year two compared to BN participants, whose average BMI increased at a modest rate (Figure 3). Additionally, participants who were older and had longer ED duration at intake had higher BMI at intake (Table 1). Note, the pattern of these results did not change when intake BMI was included in the analyses.

Table 1.

Participants’ Diagnosis at Intake, Age at Intake, and Illness Duration Predicting BMI at Intake and Intra-Individual Change in BMI in Wave 1 (N=225)

			Rate of intra-individual linear change in BMI
	BMI at intake		Intake to year 2		Year 2 to year 5		Year 5 to year 10
Predictor	B	z	B	z	B	z	B	z
BN vs. AN-R	−5.94**	−14.84	0.72**	3.54	0.27	1.73	−0.05	−0.33
BN vs. AN-BP	−4.24**	−10.52	0.59**	3.17	−0.11	−0.84	0.09	0.74
Age at intake	0.08*	2.20	−0.01	−1.21	0.01	0.05	−0.01	−0.64
Duration of ED	−0.01**	−2.78	−0.01	−1.57	0.01	0.24	0.01	0.12

Note. BMI=body mass index; BN=bulimia nervosa; AN-R=anorexia nervosa, restricting type; AN-BP=anorexia nervosa, binge-eating/purging type; ED=eating disorder.

^*p<0.05;

^**p<0.01.

Figure 3.

Shape and Rate of Intra-Individual Change in BMI from Intake to Year 2 by Intake Diagnosis

Note. BN=bulimia nervosa; AN-R=anorexia nervosa, restricting type; AN-BP=anorexia nervosa, binge-eating/purging type.

*p<0.01; **p<0.05

Aim 3: Long-term weight outcomes at 22 years

Table 2 shows mean BMI and BMI categories at intake and 22 years (Wave 2), and by recovery status at Wave 2. The majority of participants (69%) were in the normal weight range at Wave 2. A two-factor ANOVA revealed that there was a significant difference in BMI at Wave 2 among groups based on diagnostic status F(3,171)=17.37, p<0.0001, η²=0.23. Specifically, recovered participants at Wave 2 had a higher mean BMI compared to non-recovered participants with a Wave 2 diagnosis of AN-R or AN-BP (ps<0.0001), but not compared to non-recovered participants with a Wave 2 diagnosis of BN (p=0.18).

Table 2.

BMI at intake and 22 years

	BMI (kg/m2)		Underweight		Normal Weight		Overweight		Obese
	Mean	SD	N	%	N	%	N	%	N	%
Intake (N=225)	20.15	3.87	80	35.6	120	53.3	19	8.4	6	2.7
AN-R (n=42)	16.50	1.61	37	88.1	5	11.9a	0	0.0	0	0.0
AN-BP (n=77)	18.17	2.15	39	50.6	38	49.4a	0	0.0	0	0.0
BN (n=106)	23.03	3.31	4	3.8	78	73.6	20	18.9	4	3.7
22 years (N=175)	21.84	4.34	25	14.3	120	68.6	23	13.1	7	4.0
Recovered (n=113)	22.46	3.46	5	4.4	85	75.2	18	16.0	5	4.4
AN-R intake (n=25)	21.39	2.68	1	4.0	22	88.0	1	4.0	1	4.0
AN-BP intake (n=36)	21.06	2.35	4	11.1	28	77.8	4	11.1	0	0.0
BN intake (n=52)	23.94	3.87	0	0.0	35	67.3	13	25.0	4	7.7
AN-R (n=22)	18.27	3.07	11	50.0	11	50.0	0	0.0	0	0.0
AN-BP (n=13)	17.56	2.98	9	69.2	4	30.8	0	0.0	0	0.0
BN (n=13)	24.22	5.17	0	0.0	20	74.1	5	18.5	2	7.4

Note. BMI=body mass index; BN=bulimia nervosa; AN-R=anorexia nervosa, restricting type; AN-BP=anorexia nervosa, binge-eating/purging type.

^aDiagnosis at intake was initially based on DSM-III-R, before being re-assigned in 1994 to align with DSM-IV criteria. Rather than BMI cutoffs, we utilized the DSM-IV suggested weight cut-off of <85% of expected weight for AN, which do not always correspond to BMI. Note that the mean % IBW at intake was 73% (SD=9%) in the ANR group and 82% (SD=10%) in the ANBP group

Aim 4: Explaining variation in BMI at 22 years

We used a piecewise LGC model to examine how BMI at intake and change in BMI over Wave 1 (i.e., intake to year two, years two to five, and years five to ten) predicted BMI at Wave 2. Also, in the model, BMI at Wave 2 was regressed on intake diagnosis, age at intake, and illness duration. Participants’ intake BMI predicted Wave 2 BMI, B=0.69, 95% CI [0.27, 1.12], z=3.20, p<0.01. Also, participants’ rate of intra-individual change in BMI from intake to year two predicted Wave 2 BMI, B=2.02, 95% CI [0.88, 3.17], z=3.46, p<0.01. The rates of change in BMI from years two to five, B=1.81, 95% CI [−0.24, 3.87], z=1.73, p=0.08, and from years five to ten, B=1.47, 95% CI [−1.32, 4.26], z=1.03, p=0.30, were not associated with Wave 2 BMI.

To interpret the association between change in BMI from intake to year two and Wave 2 BMI, we calculated the predicted Wave 2 BMI as a function of different rates of change in BMI from intake to year two, based on estimates obtained from the LGC analysis. Predicted Wave 2 BMI was separated by intake diagnosis and calculated based on the average rate of change and the values of the lower and upper bounds of the 95% confidence interval of the average rate of change for the total sample. The predicted Wave 2 BMIs all fell within the “normal weight” range (i.e., 18.5–24.9 kg/m²) (Table 3).

Table 3.

Predicted Wave 2 BMI

		Average rate of change in BMI from intake to year 2	Predicted BMI score at 22 years
Total sample
Sample average		0.50	22.16
Lower bound of 95% CI		0.35	21.86
Upper bound of 95% CI		0.66	22.48
BN
Sample average		0.32	23.54
Lower bound of 95% CI		0.13	23.15
Upper bound of 95% CI		0.51	23.92
AN-R
Sample average		1.10	21.16
Lower bound of 95% CI		0.74	20.43
Upper bound of 95% CI		1.46	21.89
AN-BP
Sample average		0.88	21.31
Lower bound of 95% CI		0.60	20.74
Upper bound of 95% CI		1.16	21.88

Note. BMI=body mass index; BN=bulimia nervosa; AN-R=anorexia nervosa, restricting type; AN-BP=anorexia nervosa, binge-eating/purging type.

Discussion

In our sample of women seeking treatment for AN or BN, BMI increased most rapidly early in the study and plateaued over time on average in the normal weight range, with the majority of participants continuing to fall in the normal weight range 22 years later. In contrast to many patients’ predictions that they will continue to gain weight uncontrollably (8), our findings suggest that most individuals with EDs can reach and maintain a stable weight over time.

Over the course of the first ten years of the study, we found that the trajectory of BMI was initially non-linear, but eventually stabilized. Across participants, BMI increased the most during the first two years of the study, slowed down from years two to five and then leveled off from years five to ten. The shape of BMI change over time in our sample is similar to the shape of BMI change from ages 20 to 29 and ages 30 to 39 years for the 1960–1969 and 1970–1979 cohorts from the National Health and Nutrition Examination Survey (NHANES; see Figure 4 in Kranjac & Wagmiller (27)). Although we were not able to compare rate of BMI change (e.g., the change in BMI units per year) between our sample and the NHANES participants, the average BMI change for NHANES across each 9-year period was approximately 2 to 3 BMI units (27), and our sample on average changed within 2 BMI units in Wave 1. These changes are consistent with data showing that as adults grow older, increases in BMI are normative (e.g., Kranjac & Wagmiller (27); Arabshahi et al., (28)).

At 22 years after study intake (Wave 2), we found that most women who recovered or had a diagnosis of BN ultimately obtained or maintained BMIs in the normal range. A small percentage (8.2%) of those diagnosed with AN at intake, but who were recovered at Wave 2, remained in the underweight range, despite being recovered from their eating disorder. In addition, only about 20% of those who recovered were in the overweight and obese BMI range. Our findings differ from comparable population data from the 2009–2010 NHANES, which reported a 38.1% prevalence of obesity (compared to just 4.4% in our recovered sample) and a 23.2% prevalence of overweight (compared to just 16.0% in our recovered sample), among non-Hispanic white women aged 40–59 years (29).

We found that participants who had a greater rate BMI increase during the first two years of the study had higher Wave 2 BMI, regardless of their intake BMI. Of note, predicted BMI (as a function of rates of change in the first 2 years of Wave 1) was also in the normal weight range at 22 years (Wave 2). Thus, there was continued stability from Wave 1 to Wave 2. Similar to previous findings (11), participants with lower intake BMIs had slightly lower predicted BMI at 22 years than participants with higher intake BMIs. In addition, participants’ rates of BMI change between years five and ten were not associated with BMI at 22 years, which may be may be due to weight stabilization between years five and ten.

Previous research reporting longitudinal weight changes in EDs have focused on simple weight differences at follow-up time points separated by long-intervals (10, 12, 13). Our 22-year findings align with other follow-up studies that have reported BMIs in the normal weight range at 12- and 21-year follow-up (12, 13). However, no study has reported on the rate, shape, or directionality of weight change over a long-term follow-up. As the first study to date to examine the nuanced course of BMI in a sample of women with EDs for up to 22 years, our novel findings provide data that could be utilized in clinical encounters, for some individuals. Specifically, clear expectations of possible weight change during treatment can be important for patient engagement and motivation to change, particularly when fears of weight gain are unfounded or distorted.

Waller and colleagues (30) suggest that information such as weight trajectories can be utilized at the beginning of treatment to give patients a picture of what they can expect over the course of treatment, both short- and long-term. Our study provides information that can be useful for patients particularly in regard to long-term expectations. For low-weight patients (e.g., AN), weight gain is a crucial step in treatment and in our study, weight gain was steepest in the first few years after intake before stabilizing over the long-term. Although for other patients (e.g., BN) weight fluctuations can be expected as behaviors are interrupted with treatment (16), in our study overall weight change was minor for most individuals. Current cognitive-behavioral treatment for EDs (8, 31) could be supplemented with a discussion of predicted versus actual weight changes over the long-term. It will be important to examine the utility of such material in clinical practice.

Although our findings provide information that may be beneficial for treatment engagement and psychoeduction for some patients, there are others for whom fears of weight increases into the overweight or obese range may occur and other tools beyond psychoeducation are required. For example, individuals for whom weight gains or fluctuations are expected might benefit from acceptance-based strategies to reduce concerns about weight gain (32). Aligning with recent calls for research on personalized medicine (33), it will be important to understand what patient characteristics warrant what kind of education or intervention. For example, higher levels of weight suppression may result in greater rates of weight gain, in that certain individuals who are currently significantly below their highest past body weight may be fighting biological pressures to gain weight closer to their highest past weight (i.e., to achieve a homeostasis) (34–36). Future research examining predictors of weight changes and outcome, such as frequent symptom fluctuation, biobehavioral markers, and demographics, will aid in furthering our understanding of moderators and mechanisms of change. For example, determining what factors predict risk for overweight or obesity could allow for adapted interventions that include treatment targets for weight status.

Study limitations include that the sample was comprised of women diagnosed with only AN or BN and were primarily white and of higher social economic status. The study was initiated in 1987 and there may be an age-cohort effect with regard to weight, given that, more recently, individuals with BN have been increasingly presenting as overweight or obese (37). We were not able to investigate the relation between treatment and rates of change in BMI. Moreover, we were not able to contact all study participants at the follow-up assessment and thus, our conclusions must be viewed with caution. In addition, although there are controversial findings on the reliability of self-reported weight in individuals with eating disorders (38, 39), we justified reliance on self-reported weight in our sample with a large and significant correlation between baseline measured weight and baseline self-reported weight (r = .98, p < .001).

In summary, our data indicate that in a group of females with EDs, weight remained relatively stable over a longitudinal course of ten years and that, 22 years later, most were in the normal weight range. Individuals with eating disorders are often fat phobic and fear even small weight gain because of a belief that uncontrollable and large weight gain will ensue. Our evidence can be incorporated into existing interventions, in an attempt to challenge patients’ fears and facilitate treatment engagement.