Original study (Study 1)

In 1985 all 4291 individuals born in 1970 in the Swedish city of Göteborg were screened for anorexia nervosa in the eighth grade of junior high school using individual physical examinations in all cases, analysis of growth charts and an eating disorders questionnaire. ^{Reference Rastam, Gillberg and Garton10} Twenty-three girls and two boys constituted the total population of individuals born in 1970 who developed anorexia nervosa before the age of 18 years. These 25 cases correspond to a population prevalence of anorexia nervosa in those aged 17 years and under of 1.08% for girls and 0.09% for boys, and a population-corrected female:male ratio of 11.6:1. The details of the epidemiological study have been published previously. ^{Reference Rastam, Gillberg and Garton10,Reference Rastam11} One girl agreed only to examination by the school health services, leaving 24 individuals (22 girls, 2 boys) for in-depth study. Because all the teenagers in this birth cohort were examined without their clothes on, had their growth charts scrutinised in detail and were followed up for at least a year by the same school nurse, and because 99.7% of them completed an anorexia nervosa screening questionnaire, we believe that we did not miss a single case of anorexia nervosa in this cohort.

The population group was pooled with another population screening sample of anorexia nervosa cases (26 girls, 1 boy), who were reported to the research team mainly by school health nurses and doctors during follow-up of the original population group. Almost all of these individuals were born in 1971–4. This sample was less comprehensive than the original one. We estimated that the population screening sample comprised about 60% of all anorexia nervosa cases in their birth cohort. The total examined anorexia nervosa group thus consisted of 51 cases (48 girls, 3 boys) who all met or had met the DSM–III–R and DSM–IV criteria for anorexia nervosa. ^12,13

Comparison group

For each of the 51 cases of anorexia nervosa the school nurse selected the same-sex classmate closest in age to the index child. The only exclusion criteria for individuals in the comparison group were a history of eating disturbances according to school nurses' reports and growth chart deviances from age 7 years and onwards.

In the original study all 102 individuals (51 individuals in the anorexia nervosa group and 51 individuals in the comparison group) and their mothers were seen by a psychiatrist (M.R.). Within the year of assessment in the original study all 51 individuals in the anorexia nervosa group fulfilled DSM–III–R and DSM–IV criteria for anorexia nervosa, although at the precise time of examination 6 individuals in the anorexia nervosa group had bulimia nervosa and 6 further individuals at that time only partially fulfilled anorexia nervosa criteria and were diagnosed with eating disorder not otherwise specified (EDNOS) (Fig. 1). The individuals who had crossed over from anorexia nervosa to other eating disorders were not excluded from the original study or the subsequent follow-up studies.

Fig. 1

Eating disorders in the four studies of this cohort. In Study 4, only individuals with a persisting eating disorder are presented. AN, anorexia nervosa; BN, bulimia nervosa; EDNOS, eating disorder not otherwise specified; NO ED, no eating disorder.

A detailed developmental history was obtained for each child by interviewing the mother. The mothers were asked about eating disturbances from childhood onwards as well as current eating disorders. An evaluation of pregnancy, obstetrical and other medical records was made. On the basis of anonymised case notes on premorbid history, the 102 participants were assigned diagnoses of DSM–III–R personality disorders and autism-spectrum conditions by a senior psychiatrist (C.G.) who was masked to group status.

Follow-up studies (Studies 2 and 3)

All 102 individuals were examined again in Study 2 at a mean age of 21 years (6 years after anorexia nervosa onset), ^{Reference Gillberg, Rastam and Gillberg14,Reference Gillberg, Gillberg, Rastam and Johansson15} and in Study 3 at a mean age of 24 years (10 years after onset). ^{Reference Wentz, Gillberg, Gillberg and Rastam7}

Study 4

All 51 cases of anorexia nervosa were followed up in Study 4. Forty-five individuals agreed to a personal interview and five were interviewed over the telephone. In one case the mother did not want us to see her daughter, who had persisting anorexia nervosa, but agreed to a collateral interview with the semi-structured instruments used with the other participants. The 51 individuals in the comparison group all participated in Study 4. All but three were interviewed in person; these three were interviewed over the telephone. The study was approved by the human ethics committee at the medical faculty of Göteborg University. The individuals participated voluntarily after informed consent was obtained.

Measures

The Structured Clinical Interview for DSM–IV Axis I Disorders (SCID–I) and the Morgan–Russell scales were used, ^{Reference First, Gibbon and Williams16,Reference Morgan and Russell17} by either E.W. or M.R. Assessment of overall outcome was based on the three measures of the Morgan–Russell scales with the later modification by the Russell group. ^{Reference Ratnasuriya, Eisler, Szmukler and Russell3}

1. (a) General outcome: categorised as good, intermediate or poor. A good outcome is a normal body weight – a body mass index (BMI) of 18.5 kg/m² or over – and normal menstruation; an intermediate outcome is normal or near-normal body weight or normal menstruation; and poor outcome is underweight and absent or scanty menstruation.

2. (b) Persistence of disorder: individuals with overeating or vomiting (weekly or more) were assigned to the poor outcome group even if they had normal weight and menstrual cycles. ^{Reference Ratnasuriya, Eisler, Szmukler and Russell3}

3. (c) Morgan–Russell averaged scale score: this composite score summarises outcome data on body weight, diet restriction, menstruation, attitude to sex and menstruation, social relationships, relationship with family and mental state.

The Morgan–Russell instrument is currently the best-validated instrument in anorexia nervosa outcome research. ^{Reference Steinhausen1}

Statistical analysis

Scale scores were not normally distributed and therefore mainly non-parametric tests were used. For comparison between the two groups, the Mann–Whitney test was used for continuous variables and chi-squared or Fisher's exact tests for dichotomous variables. Changes over time were analysed using Wilcoxon signed rank test for continuous variables and McNemar's test for dichotomous variables. The prognostic factors used in the univariate and multivariate analyses were data obtained from before anorexia nervosa onset or before Study 2. In the univariate analysis, for comparison between the two groups, Fisher's non-parametric permutation test was used. Formal testing of correlations was performed using Pitman's non-parametric permutation test and for descriptive purposes Pearson's correlation coefficient was calculated. All dependent variables were transformed to normal distribution using Blom's normal score transformation prior to multivariate analysis, which was performed with multiple stepwise linear regression analysis. Only variables with P<0.15 in univariate tests were included as possible predictors in the multiple stepwise procedure, owing to multiple testing at each step and to possible complex interrelationships among the variables. For survival analysis, the log-rank test was used for dichotomous variables and Cox's proportional hazard model for continuous and ordered categorical variables. Survival curves were estimated with the Kaplan–Meier method. In order to select independent predictors, stepwise Cox's proportional hazard regression was performed. Only variables with P<0.15 were included as possible predictors in the multivariate analysis. Hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated for descriptive purposes. All tests were two-tailed and conducted at a 5% significance level.

Eating and other psychiatric disorders

In Study 4, six women (12%) still met the criteria for an eating disorder (Fig. 1). Altogether three women (6%) had anorexia nervosa, all with a restrictive subtype. Anorexia nervosa from onset until Study 4 was present in only one woman. One woman had been free from anorexia nervosa and other eating disorders for several years, but had had a relapse of anorexia nervosa 6 months before Study 4. Another woman had had a diagnosis of EDNOS for several years, but in Study 4 she fulfilled the criteria for anorexia nervosa. Three women had EDNOS with bulimic behaviour. Body mass index values did not differ between the anorexia nervosa and comparison groups (anorexia nervosa group, mean=22.4 kg/m², s.d.=4.5, range 13.6–37.7; comparison group, mean=23.4 kg/m², s.d.=4.2, range 18.0–38.1).

Full symptomatic recovery

Overall, 27 (54%) individuals in the anorexia nervosa group compared with 45 (88%) individuals in the comparison group had been free from disturbed behaviour and attitudes with respect to food and shape in the past 6 months (P<0.001). Eleven women (22%) in the anorexia nervosa group were ‘meat avoiders’ (i.e. they were either vegetarians (n=6) or just avoided meat (n=5)) v. one woman in the comparison group (2%) (P<0.0001). Full recovery occurred in 18% of the meat avoiders and in 63% of the ‘non-meat avoiders’ (P=0.015). Thirty-nine per cent of the anorexia nervosa group (n=20) met the criteria for at least one psychiatric disorder at follow-up (comparison group, 16%; n=8) (P=0.004) (Table 2). The single most common psychiatric disorder was obsessive–compulsive disorder, affecting 16% (n=8) in the anorexia nervosa group and none in the comparison group (P=0.006). Twenty-two per cent of the anorexia nervosa group (n=11) had an affective disorder (comparison group, 4%; n=2) (P=0.015), with a concurrent eating disorder in only two cases.

Table 2

Current and lifetime Axis I disorders in the anorexia nervosa and comparison groups

	Current diagnoses							Lifetime diagnoses
	Anorexia nervosa group					Comparison group	Anorexia nervosa group	Comparison group
	No eating disorder (n=45)	Eating disorder (n=6)	Total (n=51)	Total (n=51)	Total (n=51)	Total (n=51)
Affective disorders
Major depression
Unipolar	4	2	6	0	34	12
Bipolar I	2	0	2	1	3	1
Bipolar II	2	0	2	1	2	1
Dysthymic disorder	1	0	1	0	11	2
Any affective disorder	9	2	11 *	2	50 ***	16
Anxiety disorders
Panic disorder	1	1	2	1	8	4
Agoraphobia	0	1	1	0	1	0
Social phobia	0	1	1	0	3	1
Specific phobia	1	1	2	3	9	10
Obsessive–compulsive disorder	5	3	8 **	0	21 ***	5
Post-traumatic stress disorder	1	0	1	1	1	1
General anxiety disorder	3	2	5	1	15	8
Any anxiety disorder	10	4	14	6	37 **	20
Any anxiety disorder a	5	4	9	6	26	18
Psychotic disorders
Paranoid schizophrenia	1	0	1	1	1	1
Schizophreniform disorder	1	0	1	0	1	0
Psychosis NOS	0	0	0	0	1	0
Any psychotic disorder	2	0	2	1	3	1
Impulse control disorders
Substance misuse	0	0	0	0	2	3
Any Axis I disorder b	16	4	20 **	8	51 ***	29

NOS, not otherwise specified

a. Excluding obsessive–compulsive disorder

b. Excluding eating disorders

* P<0.05

** P<0.01

*** P<0.0001; all anorexia nervosa group v. comparison group

Predictive factors

Significant predictors and tendencies in the univariate analysis of GAF score as dependent outcome variable were early gastrointestinal problems (P=0.066), age at anorexia nervosa onset (P=0.070), autism-spectrum condition before anorexia nervosa onset and/or in Study 2 (P=0.007) and minimum BMI ever in Study 2 (P=0.043). Significant predictors and tendencies in the univariate analysis of Morgan–Russell score as dependent outcome variable were problems with friends before anorexia nervosa onset (P=0.096), premorbid obsessive–compulsive disorder (P=0.079), premorbid obsessive–compulsive personality disorder (P=0.019), age at anorexia nervosa onset (P=0.092), dysdiadochokinesia in Study 1 and/or Study 2 (P=0.033) and autism-spectrum condition before anorexia nervosa onset and/or in Study 2 (P=0.031).

Survival analysis

Table 3 shows the results for the survival analysis, using the duration of the first anorexia nervosa episode as the dependent outcome variable. The multivariate analysis rendered two significant independent predictors, namely age at anorexia nervosa onset (the lower the age, the longer the duration of the first anorexia nervosa episode) (P<0.0009; HR 1.52, 95% CI 1.19–1.94) and premorbid obsessive–compulsive personality disorder (P<0.0091; HR=0.41, 95% CI 0.21–0.80).

Table 3

Univariate survival analysis regarding duration of first episode of anorexia nervosa

Predictor	HR (95% CI)	P a
Prenatal, perinatal and neonatal factors b	1.05 (0.88-1.26)	0.58
Early gastrointestinal problems	0.99 (0.56-1.73)	0.97
Problems with friends before onset	0.76 (0.43-1.34)	0.35
Major problems in family before onset	1.33 (0.70-2.53)	0.38
Major life events before onset	1.17 (0.66-2.08)	0.58
Parental divorce before onset	1.14 (0.63-2.07)	0.67
Death of parent or sibling before onset	1.14 (0.53-2.45)	0.74
Premorbid affective disorder	1.14 (0.41-3.21)	0.80
Premorbid BMI	1.13 (1.04-1.23)	0.003
Premorbid overweight	1.41 (0.70-2.85)	0.34
Premorbid social class 1	1.25 (0.30-5.19)	0.76
Premorbid social class 2	0.76 (0.42-1.37)	0.36
Premorbid social class 3	1.27 (0.71-2.26)	0.42
Premorbid OCD	0.66 (0.33-1.31)	0.23
Premorbid OCPD	0.61 (0.33-1.11)	0.11
Age at onset	1.35 (1.08-1.69)	0.008
Primary amenorrhoea c	1.82 (0.77-4.34)	0.17

BMI, body mass index; HR, hazard ratio; OCD, obsessive–compulsive disorder; OCPD, obsessive–compulsive personality disorder

a. Cox regression model

b. Based on data from 47 individuals

c. Based on data from the 48 female participants

Predictive factors

In this study we had access to a vast amount of data for this cohort from birth onwards. In Study 1, our group collected information meticulously on environmental background factors. ^{Reference Rastam and Gillberg23} Familial and other environmental factors before anorexia nervosa onset did not turn out to be predictors of importance with respect to outcome. Low age at anorexia nervosa onset in our study predicted unfavourable outcome according to all our outcome variables, consistent with a study focusing on a sample with younger individuals. ^{Reference Bryant-Waugh, Knibbs, Fosson, Kaminski and Lask24} Premorbid obsessive–compulsive personality disorder predicted poor outcome with respect to the duration of the first anorexia nervosa episode and Morgan–Russell scores. The diagnosis of premorbid obsessive–compulsive personality disorder had been assigned by a masked rater in Study 1 based on retrospective information from a collateral interview with the mother. According to two reviews there is reasonable evidence that this disorder is associated with a poor outcome. ^{Reference Steinhausen1,Reference Crane, Roberts and Treasure25} Autistic traits occurring premorbidly and/or in late adolescence or early adult life have not previously been reported as an unfavourable prognostic factor. Autism-spectrum conditions in childhood reported in Study 1 and/or in Study 2 predicted a poor outcome according to the global functioning scales used in this study. In Study 1 and Study 2, an autism-spectrum condition was assigned by a masked rater. We believe that other researchers who have described poor social skills and obsessionality before the onset of anorexia nervosa have probably come across individuals with a disorder belonging to the autism spectrum. ^{Reference Strober, Freeman and Morrell5}

Limitations

We have assessed our sample on only four separate occasions. The retrospective data, concerning among other things relapses between studies, may be less reliable. Our sample size is relatively small in comparison with other follow-up studies. ^{Reference Theander2,Reference Strober, Freeman and Morrell5,Reference Sullivan, Bulik, Fear and Pickering26} The design of the original study, recruiting a homogeneous sample from the community and not from clinics, limited the number of available participants.

Clinical implications

To summarise, 18 years after the onset of anorexia nervosa in adolescence, the outcome regarding mortality and eating disorders was good. In terms of psychiatric and psychosocial outcomes the results were more discouraging, with 39% experiencing other psychiatric disorders and one in four unable to work. Low age at anorexia nervosa onset, obsessive–compulsive personality disorder and autistic traits predicted an unfavourable outcome.