RFS can occur within hours of commencing nutritional therapy in the severely malnourished, and this risk should be anticipated in the therapeutic plan for all patients with severe AN admitted to a medical unit (MARSIPAN 2014). There will be obvious attention to the electrolytes and to thiamine as above, but the nutritional restitution should be deliberately moderated to optimise its safety, recalling that risks exist whether patients are fed orally, enterally or parenterally. The patient with AN is not normally volume depleted unless she/he has been abusing diuretics, but if this is suspected, the circulatory volume should be carefully restored, with attention to pulse rate, blood pressure and fluid balance, with continuous cardiac monitoring in the more severe cases.

Once the hazards of the period of greatest risk of RFS have been successfully negotiated, there needs to be a clear nutritional plan based on provision of a balanced regimen of all nutrients in sufficient quantity to exceed current demands and therefore to begin the process of restoration of normal body composition. In practice this plan can be designed at the time of admission and introduced on a phased basis to a suitable percentage of the final prescription (to be expanded upon below).

At very low levels of BMI, the standard predictive equations for energy need (such as Harris–Benedict and Schofield equation) become unreliable and differ (sometimes substantially) from energy expenditure measured by indirect calorimetry. However, the great majority of units lack easy access to indirect calorimetry. A pragmatic approach must necessarily be adopted (MARSIPAN 2014).

Critical care data support a weight-based prescription, such as a final daily target of 30 kcal/kg, from which it would be hoped that basal needs would be met, together with a modest supplement for exercise and restoration of energy deficit (Singer et al. 2009). Given that the major defect in AN is starvation without a great deal of catabolic/inflammatory weight loss, this could be sufficient and indeed forms part of several published guidelines. However, there remains an important question as to “which” weight should be used, as the total will be very modest for the dramatically underweight patient if actual weight is used. There is no authoritative guidance here, although various suggestions have been made. The literature is accordingly filled with uncertainties and loose qualifying statements. In several publications previous authors have not been specific, and it has to be assumed that actual weights were intended. In some cases a fixed energy provision (such as 1,400 kcal/day) is suggested for whole groups of patients with AN, regardless of actual or ideal weight (Royal College of Psychiatrists 2004).

We looked at the data presented by Gentile et al. (2010), who performed indirect calorimetry on 33 AN patients admitted in nutritional crisis, all with a BMI of <12.5 kg/m², and where possible we have based our interpretations on the raw data provided in their paper. The mean resting energy expenditure (REE) in this patient group was 778 kcal/day, or approximately 27 kcal/kg/day, where the weight quoted was the weight of the patient on admission to the medical unit. This contrasts with the predictions from the Harris–Benedict equation of a mean REE of 1,100 kcal/day and an energy for weight of 37.8 kcal/kg/day. The patients were prescribed regimens that yielded (probably) 977 kcal/day from the first day (or 28.5 kcal/kg/day).

It is unclear what daily energy provision was subsequently given to the patients in the study by Gentile et al. (2010) as, for example, at Day 30 they are recorded as receiving 1,162 kcal orally, as well as 1,154 kcal enterally (and with an additional 194 kcal from intravenous glucose in 15 of the 33 patients). Whilst it is difficult to be sure how many calories were actually provided, the interventions were apposite and effective, as at 30 days they could record an average weight gain of 3.2 kg and a corresponding elevation of BMI by 1.3 points, with no clinical manifestations of RFS or other complications of overnutrition.

We agree with the practice of Gentile et al. to give plenty of energy and obviously also a complementary balanced input of the other nutrients, but not with their proposal to limit energy administration to correspond to the measured REE. We are convinced from the literature in general, from our own experience, and from the successful outcomes in Gentile’s practice, that energy provision in the range 25–30 kcal/kg is insufficient in the severely compromised AN patient. This conclusion is supported by the work of Garber et al. (2013), who also saw advantage in an enhanced nutritional regimen. Whitelaw et al. have also published to this effect and were unable to demonstrate any additional risk of hypophosphataemia in a carefully monitored high energy programme (Whitelaw et al. 2010). A final daily target of at least 35 kcal/kg is probably required if the weight used is the measured weight at the time of admission to the medical unit. This conclusion is obviously to be tempered by an initial restriction to prevent RFS, but one that should normally last for no more than about 10 days (see below).

We recommend caution regarding general recommendations for groups of patients of varying weight. Whilst we recommend that indirect calorimetry should be routinely performed and energy provision based on its results, we recognise that this is not practical in most centres and that there will be reliance on some form of predictive equation. The literature does not adjudicate comprehensively, and although extrapolation from closely analogous critical care practice might be reasonable, we question the hypothetical steps taken above to construct the target of 35 kcal for each kilogram of weight at admission.

Alternatively, the prescriptions could be based on the requirements of a patient of a more normal weight, as is often done in prescribing for the obese. This has the additional advantage that there is then no need to weigh the patient on admission. This is not as perverse as it may appear, as there are practical issues in weighing the acutely ill patient, and especially one who may intrinsically resent or refuse weighing, even if it is technically possible. Additionally this weight itself may be unrepresentative of nutritional status if there is fluid retention, dehydration or other important disturbance of body composition.

Pichard’s group has provided a careful summary of the means with which body composition may be assessed in AN, but they acknowledge that no technique is fully validated in this context (Mattar et al. 2011), and extrapolation to the situation of the most acutely ill is another step. Standard anthropometric measurements do not change sufficiently rapidly to be of short-term value in the medically sick patient, but grip strength may be an important exception and deserves evaluation in the MARSIPAN group of patients. If bioelectrical impedance measurement is feasible, it can provide prognostic information from the phase angle and the ratio of conductance at different frequencies (Kyle et al. 2012; Norman et al. 2012). Baseline values may be somewhat misleading in isolation but may be helpful in monitoring progress and in distinguishing weight gain that is simply fluid from that of gain in lean body mass.

Taking the approach of a prescription based on the needs of a patient of more normal weight, the body weight that would generate a nominal body mass index (BMI) of 18.5 kg/m² has been suggested. We have adjusted the measured REE values from Gentile (778 kcal/day; 27 kcal/kg/day) (Gentile et al. 2010) to the weight that would correspond to a BMI of 18.5 kg/m² (47.4 kg), which yields a new target of 1,266 kcal/day. Energy provision of 1,266 kcal equates to 26.75 kcal/kg for this semi-normalised weight. For caution and simplicity we suggest that a final target of approximately 25 kcal/kg is selected using the weight that would yield a semi-normalised BMI of 18.5 kg/m².

Both strategies—35 kcal/kg for actual weight and 25 kcal/kg for adjusted weight—should achieve weight gain without compromising the patient’s metabolic capacity. In this regard energy should be provided as both carbohydrate and fat, with an energy ratio in the region of 65:35. The frequency of hypophosphataemia seems more influenced by the initial severity of the AN than the refeeding strategy, so long as a very low threshold is maintained for phosphate repletion (Whitelaw et al. 2010).

Our practice has been to judge energy delivery primarily from nonprotein calories, on the (we believe) logical basis that we intend protein delivery as a source of new protein and not as an energy source. Whichever of the latter approaches is selected (measured current weight or idealised weight), an additional 5 kcal/kg should be added if the patient is septic or overtly catabolic.

It is at least as important to ensure adequate delivery of protein/nitrogen, and again the simplest—and usually sufficient—approach is to base delivery on the patient’s weight. The same constraints apply to interpretation of past literature as to exactly which weight that is! Recommendations to provide only 1.2 g/kg/day seem inadequate if current weight is used, and we suggest a standard final target of 1.5 g/kg/day. If a weight adjusted to a BMI of 18.5 kg/m² is used, then we recommend a protein target of 1.0 g/kg each day. We recognise that there is an arbitrary nature to the precision of our recommendations but would maintain that they are fully in line with the better validated recommendations made for general critical care practice (Singer et al. 2009) and remove the present uncertainty and probable inadequacy of past recommendations. We recognise that the adjusted weight proposals may be considered controversial but in practice have found that they yield reasonable prescriptions that have been substantially less likely to need amendment than those based on initial weight. Formal evaluation of prescribing strategy in a controlled trial setting is needed for certainty. As for energy supply (and probably even more importantly) if there are additional complications such as infection, then each of these figures should be increased.

Alongside the macronutrients there should be continuing generous provision of vitamins with particular emphasis on thiamine and ascorbate and perhaps folate. However—ironically—patients with AN quite frequently consume vitamin supplements, and major initial deficits are less common than in other patients with apparently comparable degrees of malnutrition. The trace elements also need to be considered but are usually present in sufficient quantities in ordinary food and in oral nutritional supplements, with the probable exception of selenium and zinc.

The nutritional prescription should be based as much as possible on supervised consumption of normal food. Where this is insufficient, rejected or for some other reason impracticable, then oral nutritional supplements in the form of commercially available prepackaged liquids and mousses may be helpful. It is easy to quantify the prescription but not always so easy to quantify their consumption (see below).

The limitation of energy provision is important but has been restricted differently in the various international guidelines that exist. Here we follow the general approach taken by the European Clinical Nutrition Society (ESPEN) which avoids the more extreme positions and appears to us in many ways suitably pragmatic and effective (Lochs et al. 2006). The guidance incorporates advice on nutrition and electrolyte intake and employs a phased escalation to full requirements over 4–10 days. We have however modified the recommendations to make it clear which decision has been taken on the weight of the patient and include recommendations on protein provision which curiously were omitted from the ESPEN documentation.