Eating Disorders

Steinhausen^[5**] has recently published the third in a series of comprehensive reviews of outcome studies in anorexia nervosa. The current publication spans the period 1953 to 1999 and covers 119 studies. The studies are of variable quality, and as he notes, 'few studies were prospectively organised' (p.1284). There was also 'a general lack of control conditions and a scarcity of precise information on treatment in these studies' (p.1285). But overall, he estimates that crude and standardized mortality rates were significantly high (around 5% in many studies). Of the survivors, less than one half recovered, one third improved but had not recovered, and 20% remained chronically ill. When eating and weight related symptomatology were considered in isolation, about 55% of patients made reasonable recoveries. Importantly, Steinhausen concluded that there was no convincing evidence that the outcome of anorexia nervosa improved over the second half of the 20th century. It seems that the figures represent the natural history of anorexia nervosa after contact with treatment services, and that little that has occurred in the last 50 years has influenced that history.

The Issue of Exposure

The majority of studies accessed by Steinhausen^[5**] are follow-through studies of patients after they have completed a treatment programme of some kind. It appears that this method has been employed to support the continued provision of comprehensive and lengthy treatment programmes in many countries, including both Germany and the Netherlands.^[6**] Such arguments fly in the face of established epidemiological reasoning. It comes down to a matter of exposure.

If a whole population is exposed to a constant dose of an agent such as an inpatient treatment programme, the future status of that population cannot simply be assumed to be consequent on that exposure. Whatever happens might have happened anyway, exposure or not. The best way of demonstrating that an exposure truly has an effect is to find a lawful relationship between variations in intensity of exposure and degree of response, even if that exposure is all or nothing, as in most placebo-controlled treatment trials. Such lawful variations are embodied in the notion of dose-response patterns. Demonstrating their existence is an essential element in the attribution of causality either to an environmental agent thought to cause harm, or to an agent such as a treatment programme that is assumed to be helpful. The post-hoc nature of most published outcome studies, the frequency with which patients have been exposed to a uniform treatment package, or the lack of specificity about variations in treatment intensity where such variations exist, led Steinhausen^[5**] to conclude that the studies he collected could not be used for a 'definitive evaluation of treatment effects' (p.1285). This is a doleful assessment of generations of effort.

Our recent study^[7**] of the outcome of patients with eating disorders, conducted in South Australia, was designed so as to conform to basic epidemiological principles; its conclusions warrant closer study.

The South Australian Study of the Outcome of Eating Disorders

The study by Ben-Tovim and colleagues^[7**] was conducted in Adelaide in South Australia, a city whose characteristics make it well suited to epidemiological investigations. The nearest major population centre is over 700 km away, so that the full range of specialized medical treatments have to be provided within the city. The population is a little over a million people. This is large enough to support a range of treatment facilities for patients with eating disorders and small enough for the specialist practitioners to know and cooperate with each other, which they did in this study. Over a 20-month period, consecutive referrals to all Adelaide-based secondary and tertiary eating disorders treatment services (including an ex-sufferer-ran alternative treatment service) were recruited at the point of referral. The final study sample included 95 patients with anorexia nervosa, 87 with bulimia nervosa, and 37 with eating disorder not otherwise specified (EDNOS). Ninety-eight per cent of patients were followed up and re-studied 5 years later. This was a study of a representative group of eating disordered patients recruited at the point of specialist referral, the period of interest for care-providers.

What makes this study^[7**] of particular interest is that the patients were recruited and studied independently of any treatment that they received. That treatment varied from prolonged specialized inpatient treatment to no specialist treatment for six anorexic and 20 bulimic patients, or very minimal such treatment comprising no more than three outpatient sessions over the study period for a further 20 anorexic and 35 bulimic patients. The variability was a consequence of patient preference rather than rationing of care. Existing resources were sufficient to provide the range of treatments for all that sought them, and there were no restrictions imposed by third parties such as insurance companies or managed care providers.^[8*]

The Outcome of Anorexia Nervosa

The outcome for the anorexic patients was in line with the overall findings in the Steinhausen review.^[5**] Two of the patients died as a direct consequence of their condition (as did two of the EDNOS patients). 59% had no diagnosable eating disorder at 5 years, though when psychosocial functioning was taken into account only one third of patients made a complete recovery. A worse outcome was predicted by a lower body weight, more frequent binge-vomiting at presentation, and less rapid change in body related attitudes,^[9*] but not by a longer duration of illness before presentation.

The varying intensity of treatments accessed by the patients would have made it possible to demonstrate a dose-response relationship between treatment and outcome, but no relationship between the intensity of treatments and eventual outcome could be identified. That is, once the patient outcomes had been adjusted for the base-line variation in those initial characteristics that independently influenced outcome, no further information about the final outcome was added by knowing about the differing treatments that the patients underwent. So, no relationship could be established between outcome and exposure (and lack of exposure) to treatment after initial presentation. We concluded that 'the efficacy of existing treatment is questionable' (p.1254).

This was an observational study, and due caution should be attached to its conclusions. A naturalistic study cannot compare in robustness with a randomized treatment trial. But in anorexia nervosa, there is no substantial treatment literature against which to put our study. The primary author of one of the few random allocation studies of primary treatment in anorexia nervosa^[10] has recently written in heartfelt terms about the difficulties of that study.^[11**] Its complex design,^[12] with its attempts to allocate patients to four treatment groups, meant that the numbers recruited were always likely to be too small to demonstrate the effectiveness of any one treatment arm, and the study foundered after 1 year of follow-up. Indeed extrapolation of the figures from the Ben-Tovim study indicates that between 200 and 300 participants are needed to be confident about being able to draw conclusions from a treatment trial in which even one active treatment is compared against a control treatment. With the disarming frankness afforded by retirement, Crisp^[11**] writes that whilst resource intensive inpatient care provides a 'firm and substantial funding base in private health care systems and lends gravitas to the speciality (p.238)' he is less committed to it being a necessary part of treatment itself. Hardly surprising then that in the world of managed care, third-party payers are increasingly reluctant to support such treatments.^[13*]

The Outcome of Bulimia Nervosa and EDNOS

Our findings in relation to bulimia nervosa were much more hopeful. Almost three quarters of the bulimic participants had no diagnosable eating disorder at 5 years, and had made a good overall recovery, figures very much in line with the best long term outcomes from established treatment programmes.^{[14, 15]} The increased likelihood of final recovery was predicted by better initial psychosocial functioning and less intense negative body related attitudes at presentation.

The differential outcomes of anorexia and bulimia nervosa, and the differential predictors of those outcomes, argue for a clear separation between the conditions, rather than placing them along some kind of continuum.^[16*] Whilst two EDNOS patient had died during the course of the study, the majority made a good recovery. It was not possible to identify predictors of EDNOS outcome, and the robustness of EDNOS as a diagnostic category remains problematic.

As with anorexia, it was not possible in the South Australian study to relate treatment to outcome in bulimia nervosa. The majority of bulimic patients received no or very minimal specialist treatment during the 5-year study period, yet their overall outcome was excellent.

Substantial treatment studies of bulimia nervosa have begun to appear. One of the best designed of those studies compared two treatments against each other, with no placebo or control treatment arm.^[17] Such a design is only justifiable if it is clear that the condition under consideration will not improve in the absence of either of the treatments provided. The South Australian study undermines any such assumption. It is just not possible to be confident^[6**] that treatments such as cognitive-behavioural or interpersonal psychotherapy influence long-term outcomes in bulimia nervosa. It may be that the act of seeking specialist assistance sets in train a natural process of remission that subsequent treatment only appears to influence. In which case, treatment trials which compare treatments against each other and conclude that all treatments are equally effective can also be interpreted as demonstrating that none has any specific effect and that all that is being observed is a natural process of remission. Put another way, the South Australian study implies that a treatment trial of bulimia nervosa without a placebo or minimal treatment arm is functionally uninterpretable, at least in relation to outcome in the long term. It may be that the more appropriate goal of such treatment trials is to determine whether treatments diminish total periods of disability, rather than influencing eventual recovery.

Prevention

If the benefits of treatment are uncertain, what of prevention? Existing prevention programmes have used one of two approaches; in a primary prevention strategy, an attempt is made to 'inoculate' individuals by strengthening their natural defences against what are assumed to be primary environmental causative agents such as unrealistic media representations of body shapes or socially sanctioned inappropriate eating related behaviours. A secondary prevention or early intervention approach is to seek out emerging manifestations of disorders, and by intervention, reduce the likelihood of disease progression.

Eating disorders of clinical severity are relatively rare. The prevalence of anorexia nervosa is around 0.5% of the adolescent and early adult female population, and the prevalence of bulimia nervosa between 1 and 2%. These numbers are too small to allow secondary prevention programmes to be feasible in any other than a high risk population,^[18*-20*] and as we have seen, the efficacy of any provided intervention is itself likely to be in doubt. So most groups have favoured a primary intervention approach using an educational package as the vaccinating^[21*] agent. The inoculation is mostly undertaken in schools or other environments where it is possible to provide the intervention in a cost-efficient manner.

Unfortunately, a recent comprehensive review^[22**] undertaken under the aegis of the Cochrane collaboration casts serious doubts on the value of such an approach. Whilst the primary prevention programmes reviewed did not appear to harm the participants, there was no substantial evidence that they provided sustainable benefits in relation to eating disorder behaviours or attitudes thought to influence the emergence of eating disorders. This is hardly surprising.

In epidemiological terms, contemporary Western culture is a universal exposure. Whatever its merits, all women in Westernized societies are exposed to it, and the forces that maintain that culture are clearly very powerful. Attempting to change a whole culture or to inoculate against its effects by means of a brief prevention programme was always likely to be difficult. Whilst that difficulty is not a reason to avoid trying, campaigners are hindered by our lack of understanding of the precise factors that initiate disorders of clinical significance.^{[23**, 24**, 25*]}

Clinically severe eating disorders are relatively rare. Simple arithmetic indicates that a universal factor cannot alone be causative of a rare condition; otherwise the whole population would be affected. It can only increase the background risk in the most general of ways. Westernized culture may or may not be necessary for the development of a rare disorder, but it certainly cannot be sufficient. There must be more condition specific initiating factors that generate onsets of relatively rare disorders. The existence of such factors cannot be in doubt, only their nature. Existing theories of the aetiology of eating disorders lack specificity about initiating factors, and until those factors, be they biological, social or psychological, are identified, prevention campaigns are unlikely to be successful.

Conclusion

This is an evidential era. Authorities are becoming increasingly sceptical about supporting treatment and prevention programmes whose benefits, not demonstrated by reference to objective evidence, can only be asserted by their proponents. Prevention will only be possible when specific and reversible factors that lead to the initiation of eating disorders have been identified. However, treatment is often provided on an empirical basis. Here, the most pressing need is for a large-scale study comparing an intensive and comprehensive inpatient weight restoration programme for patients with anorexia nervosa against outpatient supportive care plus access to brief hospitalization for nutritional stabilization. Such a study is likely to need between 200 and 300 participants, and so would require cooperation between several centres. But without clearer evidence of benefit, the future of apparently expensive inpatient treatment programmes will remain in doubt.

Reprint Address

Correspondence to David I. Ben-Tovim, Clinical Epidemiology Unit, Flinders Medical Centre, Bedford Park, South Australia 5042 e-mail: david.ben-tovim @fmc.sa.gov.au

Abbreviation Notes

EDNOS, eating disorder not otherwise specified

David I. Ben-Tovim, Clinical Epidemiology Unit and Department of Psychiatry, Flinders Medical Centre, Bedford Park, South Australia