Prediction equations

Measurements of energy expenditure by direct or indirect calorimetry or the doubly-labelled-water method are the most accurate methods for determining energy requirements. In the clinical situation in the UK, however, these methods are usually impractical because they are expensive, time-consuming and require trained personnel to perform them. As a result, a considerable number of prediction equations have been published over the past 40 years in an attempt to develop more practical tools for determining energy requirements in the clinical setting. While usually quick and easy to use, inexpensive and universally available, all prediction equations are open to criticism for a number of reasons. All the equations require an assessment of current weight and they have not been adequately validated. While the equations may accurately predict energy requirements for specific populations, they have a poor predictive value for individuals. In addition, all methods require some extent of clinical judgement and are therefore open to misinterpretation.

The basis of several prediction methods is an estimation of BMR (Wilmore, Reference Wilmore1977; Long et al. Reference Long, Schaffel, Geiger, Schiller and Blakemore1979; Elia, Reference Elia1990; Todorovic & Micklewright, Reference Todorovic and Micklewright2004) to which is added a ‘stress’ or ‘injury’ factor to take account of the changes in energy expenditure presumed to have resulted from illness or injury. During the early part of the 20th century BMR measurements such as those conducted by Harris & Benedict (Reference Harris and Benedict1919; HBE) were primarily used to diagnose hypo- and hyperthyroidism. It was not until the 1980s that a publication by the Food and Agriculture Organization/World Health Organization/United Nations University (1985) used measurements of energy expenditure (including BMR), rather than food intake, to estimate energy requirements. In the USA the most-commonly-used BMR prediction equations are those developed by Harris & Benedict (Reference Harris and Benedict1919; HBE), yet they are open to criticism. The equations were developed from measurements conducted on 136 males and 103 females over a 10-year period between 1909 and 1917, about one century ago. The subjects were considered healthy and ‘normal’ for that time, but compared with current populations were young (females, 27 (sd 9) years; males, 31 (sd 14) years) and lean (BMI: females, 21·4 (sd 2·8) kg/m²; males, 21·5 (sd 4·1) kg/m²). The predictive value of the HBE for young females is poor, a fact later recognised by Benedict himself (Benedict, Reference Benedict1928), and more recent studies comparing the HBE with measured BMR in healthy populations show that the HBE consistently overestimate by 5–15% (Daly et al. Reference Daly, Heymsfield, Head, Harvey, Nixon, Katzeff and Grossman1985; Mifflin et al. Reference Mifflin, St Joer, Hill, Scott, Daugherty and Koh1990). The Schofield equations were developed more recently as the basis for the Food and Agriculture Organization/World Health Organization/United Nations University (1985) report and are commonly used in Europe and other parts of the world. The equations were developed from a meta-analysis of about 100 studies conducted over a long time period (1914–1980), including the subjects studied by Harris & Benedict (Reference Harris and Benedict1919). However, these equations have also been shown to overestimate BMR in many contemporary populations (Henry, Reference Henry2005). This disparity is in part because the database is skewed towards a younger population, with only eighty-eight subjects (1·2% of the total subjects) aged >60 years. Furthermore, the database contains a disproportionate number of Italian subjects (3388 of 7173; 47%), who were found to have a higher BMR on a per kilogram body weight basis than any of the other study populations; possibly because two-thirds of the subjects (2279 of 3388; 67%) were young men who were leading physically-active lives, either in military service or employed as miners and labourers. Recently, a series of newer BMR prediction equations has been developed (Henry, Reference Henry2005). These equations take account of many of the earlier criticisms and they tend to produce lower BMR values than the current Food and Agriculture Organization/World Health Organization/United Nations University (1985) equations. These newer equations have yet to be fully validated, and their future use and application will depend on their ability to predict BMR more accurately in contemporary populations. Whichever BMR prediction equation is used in the clinical situation, it must be noted that all equations have been derived from healthy populations and therefore may not be an ideal baseline for estimating requirements in sick or injured individuals.

In addition to details on age and gender these BMR equations require a current weight. While it is possible to obtain a reliable weight in many hospitalised patients, especially with the increasing availability of chair, hoist and bed scales, achieving an accurate weight still remains problematic in a considerable number of patients, in particular those with impaired balance or mobility, patients with open wounds and those attached to medical equipment such as ventilators. Furthermore, an increasing number of patients are being fed by artificial nutrition support at home, yet many are bed-bound or immobile and therefore difficult to weigh. Fluid retention also makes it difficult to assess true body weight. In critically-ill patients the accumulation of 10–20 litres extracellular fluid may occur during the acute phase of injury (Lobo et al. Reference Lobo, Macafee and Allison2006) and in patients with liver disease similar amounts of fluid may accumulate as ascites. Even the ankle oedema commonly seen in malnutrition and in chronic diseases such as chronic obstructive pulmonary disease may make a major contribution to a patient's body weight. However, fluid retention increases body weight without increasing metabolically-active tissue and therefore, where possible, ‘dry’ weight should be recorded and used to calculate BMR. For example, post-dialysis weight should be used in renal patients or weight following drainage of ascites in liver disease (Madden & Wicks, Reference Madden and Wicks1994). In patients in the ITU who have undergone surgery it has been shown (Weissman & Kemper, Reference Weissman and Kemper1992) that subtracting the weight of cumulative net fluid balance from post-operative weight accurately reflects preoperative weight. In the absence of hoist or bed scales in the ITU it is common practice for clinicians to estimate a patient's weight and use this estimate for calculating energy requirements and drug doses. This method relies on clinical judgement and, to date, there appear to be no studies using clinician-estimated weight compared with, for example, last known weight, in relation to estimating energy requirements in patients in the ITU. In conscious and orientated patients self-reported weight can be used as a surrogate for measured weight, although it is less reliable in obese individuals who tend to underestimate their weight and lean individuals who tend to overestimate their weight (Rowland, Reference Rowland1990; Roberts, Reference Roberts1995; Elia, Reference Elia2003; Ruivo et al. Reference Ruivo, Marques-Vidal and Camilo2004).

In a recent review Reeves & Capra (Reference Reeves and Capra2003) have attempted to determine the evidence-base of a number of current prediction methods (Wilmore, Reference Wilmore1977; Long et al. Reference Long, Schaffel, Geiger, Schiller and Blakemore1979; Elia, Reference Elia1990; Todorovic & Micklewright, Reference Todorovic and Micklewright1997). The authors conclude that the original data on which the methods were based are difficult to locate, that the methods tend to be based on expert opinion or consensus statement rather than experimental evidence and that there is a lack of information on how the methods were derived. Furthermore, none of the methods has been adequately validated.

A number of studies have compared a variety of prediction equations with measured energy expenditure (MEE) in different ITU populations (Carlsson et al. Reference Carlsson, Nordenstrom and Hedenstierna1984; van Lanschot et al. Reference van Lanschot, Feenstra, Vermeij and Bruining1986; Weissman et al. Reference Weissman, Kemper, Elwyn, Askanazi, Hyman and Kinney1986; Swinamer et al. Reference Swinamer, Phang, Jones, Grace and King1987; Hunter et al. Reference Hunter, Jaksic, Lewis, Benotti, Blackburn and Bistrian1988; Cortes & Nelson, Reference Cortes and Nelson1989; Casati et al. Reference Casati, Colombo, Leggieri, Muttini, Capocasa and Gallioli1996; Flancbaum et al. Reference Flancbaum, Choban, Sambucco, Verducci and Burge1999; Cheng et al. Reference Cheng, Chen, Wong, Lee, Kan and Huang2002). While the majority have found that prediction equations tend to overestimate requirements compared with MEE and that all equations have a poor predictive value for individuals, it was not always the case. This inconsistency may in part be because some investigators have compared MEE with a BMR equation such as HBE (Weissman et al. Reference Weissman, Kemper, Elwyn, Askanazi, Hyman and Kinney1986; Swinamer et al. Reference Swinamer, Phang, Jones, Grace and King1987; Hunter et al. Reference Hunter, Jaksic, Lewis, Benotti, Blackburn and Bistrian1988), others have compared MEE with predicted BMR plus variable stress and/or activity factors (Carlsson et al. Reference Carlsson, Nordenstrom and Hedenstierna1984; van Lanschot et al. Reference van Lanschot, Feenstra, Vermeij and Bruining1986; Cortes & Nelson, Reference Cortes and Nelson1989; Casati et al. Reference Casati, Colombo, Leggieri, Muttini, Capocasa and Gallioli1996; Cheng et al. Reference Cheng, Chen, Wong, Lee, Kan and Huang2002) and others have compared MEE with equations derived specifically for ITU populations (Flancbaum et al. Reference Flancbaum, Choban, Sambucco, Verducci and Burge1999; Cheng et al. Reference Cheng, Chen, Wong, Lee, Kan and Huang2002). None of the studies has given specific details of the clinical variables used to assign the stress factors, nor have they commented on the profession, knowledge and experience of the individual calculating the predicted requirements.

Prediction equations that state a range (for example, see European Society for Parenteral and Enteral Nutrition guidelines; Kreymann et al. Reference Kreymann, Berger, Deutz, Hiesmayr, Jolliet, Kazandjiev, Nitenberg, van den Berghe and Wernerman2006) and those that require the use of stress factors (for example, see Parenteral and Enteral Nutrition Group of the British Dietetic Association guidelines; Todorovic & Micklewright, Reference Todorovic and Micklewright2004) rely on clinical judgement, and may therefore be open to misinterpretation, especially in those users who are unfamiliar with the evidence-base (Glynn et al. Reference Glynn, Greene, Winkler and Albina1999; Barak et al. Reference Barak, Wall-Alonso and Sitrin2002; Reeves & Capra, Reference Reeves and Capra2003). During the past three decades prescribed energy requirements for critically-ill patients have been revised downwards, in part because of changes in the measurement and interpretation of energy expenditure and in part as a result of changes in therapeutic techniques. In early studies measurements of energy expenditure were frequently made close to the time of peak hypermetabolism (i.e. the first few days of injury or illness) and then extrapolated, inappropriately, to much longer periods (Fig. 2). For example, Vermeij et al. (Reference Vermeij, Feenstra, van Lanschot and Bruining1989) have found that when TEE on day 1 is used to estimate TEE on subsequent days in sixty patients who have undergone surgery and are mechanically ventilated errors of ⩽31% of actual TEE occur. Also, reported measurements were often conducted while patients were receiving ‘hyperalimentation’, i.e. large amounts of parenteral nutrition (either continuously or intermittently), and therefore a sizable proportion of dietary-induced thermogenesis may have been included in the measurement (Elia & Jebb, Reference Elia and Jebb1992).

Fig. 2.

The effects of the metabolic response to injury on energy expenditure (kJ/d).

A number of therapeutic interventions have been shown to either increase or decrease energy expenditure. For example, in the past it was the practice to attempt to cool patients with pyrexia by lowering ambient temperature. However, this practice caused an increase in thermogenesis as the body attempted to maintain a high core temperature. More recently, heat loss, and therefore energy expenditure, has been minimised by nursing patients at a higher ambient temperature, i.e. in a thermo-neutral environment (Elwyn et al. Reference Elwyn, Kinney and Askanazi1981). Other changes that have reduced both the magnitude and duration of the stress response in the past two decades include: the elective and early surgical removal of necrotic tissue, drainage of abscesses, the early diagnosis and aggressive management of infections (Elia, Reference Elia1995); reduction in the use of corticosteroids in closed head injury (Greenblatt et al. Reference Greenblatt, Long, Blakemore, Dennis, Rayort and Geiger1989); use of ibuprofen in burn injury (Wallace et al. Reference Wallace, Caldwell and Cone1992); re-warming of patients following coronary-artery bypass surgery (Hanhela et al. Reference Hanhela, Mustonen, Korhonen and Salomaki1999); the use of effective sedation and paralysis in patients who are mechanically ventilated (Bruder et al. Reference Bruder, Raynal, Pellissier, Courtinat and Francois1998; Terao et al. Reference Terao, Miura, Saito, Sekino, Fukusaki and Sumikawa2003).

Some of the prediction equations (Todorovic & Micklewright, Reference Todorovic and Micklewright2004) require the clinician to add a factor for activity, yet the factor assigned is dependent on clinical judgement. The effects of physical activity on TEE depend on tissue mass and type of activity, its intensity, duration and frequency, yet in routine clinical practice these variables are rarely, if ever, formally assessed. While there are recommendations for activity factors for use in hospitalised patients (Elia, Reference Elia1990), the source data for these guidelines are difficult to locate. In the ITU setting activities such as weighing on a sling-type bed scale, repositioning and chest physiotherapy result in an increase in energy expenditure of 20–30%. However, the effect is transient, lasting ⩽30 min post intervention, and therefore the contribution of such activities to TEE is likely to be small, i.e. 5–10% (Weissman et al. Reference Weissman, Kemper, Elwyn, Askanazi, Hyman and Kinney1986; Swinamer et al. Reference Swinamer, Phang, Jones, Grace and King1987). A literature review reveals a paucity of studies measuring the effects on physical activity (and thus energy expenditure) of abnormal neuromuscular function such as motor neurone disease, passive and active physiotherapy, the increased effort involved in moving injured and/or painful limbs and mechanical inefficiency, e.g. secondary to chronic obstructive pulmonary disease. The few studies available, e.g. in cerebral palsy (Johnson et al. Reference Johnson, Hildreth, Contompasis and Goran1997), heredity neuro-muscular disease (McCrory et al. Reference McCrory, Kim, Wright, Lovelady, Aitkens and Kilmer1998), Parkinson's disease (Toth et al. Reference Toth, Fishman and Poehlman1997) and hip fracture (Miller et al. Reference Miller, Daniels, Bannerman and Crotty2005), suggest that the effects on TEE are very variable and require further investigation. Furthermore, an increasing number of patients are being discharged home on artificial nutritional support and, while a large proportion is house-bound, there are a number of patients who regularly leave the house to socialise or even work, albeit part-time. While the physical activity levels in the Department of Health (1991) report may apply to some patients in the latter group, there is a need to conduct more research on the effects of physical activity TEE in different clinical conditions.

An alternative simpler method for determining energy requirements is one based on set energy values expressed on a per kg body weight basis. For example, recent recommendations include 84–105 kJ (20–25 kcal)/kg body weight for critically-ill patients (Kreymann et al. Reference Kreymann, Berger, Deutz, Hiesmayr, Jolliet, Kazandjiev, Nitenberg, van den Berghe and Wernerman2006) and 105–146 kJ (25–35 kcal)/kg body weight for patients who are not severely ill or injured, nor at risk of re-feeding syndrome (National Institute for Health and Clinical Excellence, 2006). Other authors have recommended different cut-offs for the same disease states (American Society for Parenteral and Enteral Nutrition Board of Directors, 2002). The method appears to have been originally derived for critically-ill patients who are mechanically ventilated, yet explanations of how the values were derived or references to original work are difficult to locate and there appear to be no studies validating this method in critically-ill or other patient groups. It is unclear if the method is valid for use in underweight or obese subjects and, if it is, whether clinicians should use actual, ideal or adjusted body weight. Furthermore, the method does not take account of differences in energy expenditure resulting from differences in age or gender. In addition to a lack of validation, there appears to be no guidance on how to determine where in the range a particular patient might be placed. For example, what variables should be used to determine whether a metabolically-stable patient post surgery with a BMI in the acceptable range (20–25 kg/m²) receives 84 (20), 105 (30) or 146 (35) kJ (kcal)/kg body weight per d? Currently, the level used depends on clinical judgement and the experience and knowledge of the practitioner. Studies in ITU populations have shown that REE on a per kg body weight basis can vary considerably from as low as 42 kJ (10 kcal) to >209 kJ (50 kcal; Frankenfield et al. Reference Frankenfield, Omert, Wiles, Bagley, Goodarzi and Siegel1994; Green et al. Reference Green, Campbell, McClelland, Hutton, Ahmed and Helliwell1995; McClave et al. Reference McClave, Lowen, Kleber, Nicholson, Jimmerson, McConnell and Jung1998; Reid & Campbell, Reference Reid and Campbell2001;Barak et al. Reference Barak, Wall-Alonso and Sitrin2002), and in patients with sepsis the reported ranges are even greater, varying from 60 kJ (14·4 kcal)/kg body weight to 364 kJ (87 kcal)/kg body weight (Shizgal & Martin, Reference Shizgal and Martin1988; Frankenfield et al. Reference Frankenfield, Omert, Wiles, Bagley, Goodarzi and Siegel1994; Uehara et al. Reference Uehara, Plank and Hill1999). These studies illustrate that assigning a set value on a per kg body weight basis, e.g. 84 kJ (20 kcal), may result in marked under- or overfeeding in a proportion of patients in the ITU.

Measured energy expenditure

The estimation of energy requirements is so challenging in some conditions, e.g. critical illness, obesity and liver disease, that it is recommended that expenditure be measured on an individual basis by indirect calorimetry (American Society for Parenteral and Enteral Nutrition Board of Directors, 2002). However, the equipment is not readily available in the UK, and there are a number of technical and practical considerations that need to be taken into account to ensure accurate and reliable measurements, recently summarised by Branson & Johannigman (Reference Branson and Johannigman2004). Conditions should be standardised so that comparisons with future measurements under the same conditions will be meaningful. During measurements it is therefore important to prevent interaction between patient, healthcare professionals or visitors and to avoid procedures that will influence accuracy, e.g. haemo-dialysis or filtration. It is difficult to determine how long measurements should continue, although achieving a steady-state (i.e. change in V_O2 of <5% from 1 min to the next) improves the validity of measurements (Brandi et al. Reference Brandi, Bertolini and Calafa1997; McClave et al. Reference McClave, Spain, Skolnick, Lowen, Kieber, Wickerham, Vogt and Looney2003). In metabolically-stable patients it has been shown that a steady-state period of 5 min (Petros & Engelmann, Reference Petros and Engelmann2001) or 2×15 min is sufficient to predict 24 h REE (Behrendt et al. Reference Behrendt, Surmann, Raumanns and Giani1991). In metabolically-unstable patients, however, longer periods of from 30 min to 2 h may be required (Smyrnios et al. Reference Smyrnios, Curley and Shaker1997; Branson & Johannigman, Reference Branson and Johannigman2004). Generally, clinically-stable patients will require shorter and less-frequent measurements than those who are more unstable, e.g. those with spiking pyrexia or haemodynamic instability and those in the immediate post-operative state (Weissman et al. Reference Weissman, Kemper and Hyman1989). While it may seem an appropriate variable, it appears that injury severity score and other measures of severity of illness are not reliable as predictors of energy requirements in patients in the ITU. Several studies have reported that REE cannot be predicted by the injury severity score or the APACHE II severity of illness score (Vermeij et al. Reference Vermeij, Feenstra, van Lanschot and Bruining1989; Rodriguez et al. Reference Rodriguez, Sandoval and Clevenger1995; Boulanger et al. Reference Boulanger, Nayman, McLean, Phillips and Rizoli1994; Brandi et al. Reference Brandi, Bertolini, Santini and Cavani1999), although it was not found to be the case in other studies (Swinamer et al. Reference Swinamer, Phang, Jones, Grace and King1987; Brown et al. Reference Brown, McClave, Hoy, Short, Sexton and Meyer1993; Hwang et al. Reference Hwang, Huang and Chen1993).

While the majority of patients in the ITU will leave the unit within 7 d, some patients may have a protracted stay as they suffer sequential failure of various organ systems. To date, very few studies have measured energy expenditure during prolonged ITU stays. Monk et al. (Reference Monk, Plank, Franch-Arcas, Finn, Streat and Hill1996) have measured REE in ten critically-injured patients as soon as they were haemodynamically stable and then every 5 d for 21 d. In these patients TEE was found to rise to 1·55×REE on day 10 and to remain elevated throughout the study period. The same group (Plank et al. Reference Plank, Connolly and Hill1998) subsequently conducted a similar study on twelve patients with severe sepsis secondary to peritonitis and have reported that TEE rises to 1·25×REE and again remains elevated throughout the study period. This area is one that requires further investigation.

In the ITU setting, therefore, patients who may benefit from indirect calorimetry measurements include those who fail to respond adequately to estimated nutritional requirements and patients with multi-organ failure who require prolonged ventilatory support (Brandi et al. Reference Brandi, Bertolini and Calafa1997), in addition to very-underweight or obese patients. Outside the ITU setting many patients require nutrition support over prolonged periods of several months or even years, yet there is a relative lack of studies that report MEE in such patients, partly because of the lack of affordable portable equipment that can record energy expenditure over several days or weeks. Until recently, long-term measurements could only be achieved through the use of the doubly-labelled-water technique, which is both expensive and requires considerable expertise. However, recently-launched products, such as the Sensewear^® armband (BodyMedia, Pittsburgh, PA, USA), may allow researchers to answer some of these questions in the future, but currently they require more validation before their use can be recommended for routine clinical practice.

Clinical implications

While it is possible to measure energy expenditure in the clinical situation, it is important to note that MEE does not necessarily equate to energy requirements. For example, if a patient in the ITU is measured while fasting and lying still MEE may underestimate TEE, as no factors have been included for dietary-induced thermogenesis or activity. Furthermore, as described earlier, energy requirements are affected by a multitude of factors that are not included in prediction equations or taken into account during measurements of energy expenditure. Examples of these factors include the goals and likely duration of nutritional support and the specific requirements of obese patients.

In the patient in the ITU meeting energy requirements does not prevent loss of lean body mass, and exceeding requirements causes metabolic complications and may precipitate organ dysfunction leading to increased ventilator dependence and length of ITU stay (Kinney & Elwyn, Reference Kinney and Elwyn1985; Streat et al. Reference Streat, Beddoe and Hill1987; Frankenfield et al. Reference Frankenfield, Smith and Cooney1997; Hart et al. Reference Hart, Wolf, Herndon, Chinkes, Lal, Obeng, Beauford and Mlcak2002; Plank & Hill, Reference Plank and Hill2003). There is therefore a question of what are the goals of nutritional support in the ITU setting. It would seem prudent to aim to minimise losses during the acute phase of illness but then to ensure adequate repletion in the recovery (anabolic) phase. A major problem with this approach, however, is that currently there are no universally-agreed definitions of under- and overfeeding.

Energy intake >110% estimated requirements has been used as a definition by some authors (McClave et al. Reference McClave, Lowen, Kleber, Nicholson, Jimmerson, McConnell and Jung1998; Alberda et al. Reference Alberda, Snowden, McCargar and Gramlich2002; Kan et al. Reference Kan, Chang, Sheu, Cheng, Lee and Huang2003); however, the requirements have been estimated differently by the different authors. For example, McClave et al. (Reference McClave, Lowen, Kleber, Nicholson, Jimmerson, McConnell and Jung1998) and Alberda et al. (Reference Alberda, Snowden, McCargar and Gramlich2002) have estimated requirements by adding 10% to REE measured by metabolic cart, whereas Kan et al. (Reference Kan, Chang, Sheu, Cheng, Lee and Huang2003) have estimated the requirements by adding 20% to REE. As a result of the differences in how the energy requirements have been estimated, there are considerable differences in the reported rates of ‘overfeeding’. Furthermore, in a study comparing practice in thirty-two hospitals a variation between centres was found (McClave et al. Reference McClave, Lowen, Kleber, Nicholson, Jimmerson, McConnell and Jung1998) in the percentage of patients being overfed, from 32·2 to 92·8, highlighting considerable differences in practice.

In relation to underfeeding, it is important to make the distinction between hypoenergetic feeding (low energy with adequate protein) and underfeeding (low energy and protein). There is some evidence that hypoenergetic feeding may be associated with better outcomes in the ITU (Ibrahim et al. Reference Ibrahim, Mehringer, Prentice, Sherman, Schaiff, Fraser and Kollef2002; Krishnan et al. Reference Krishnan, Parce, Martinez, Diette and Brower2003), especially in patients receiving parenteral nutrition (Patino et al. Reference Patino, de Pimiento, Vergara, Savino, Rodriguez and Escallon1999) and obese patients (Burge et al. Reference Burge, Goon, Choban and Flancbaum1994; Choban et al. Reference Choban, Burge, Scales and Flancbaum1997; Dickerson et al. Reference Dickerson, Boschert, Kudsk and Brown2002). This type of outcome may not be achieved, however, in patients who are not obese (Villet et al. Reference Villet, Chiolero, Bollmann, Revelly, Cayeux, Delarue and Berger2005). Furthermore, underfeeding, leading to marked negative N balance during ITU stay, may result in poorer outcomes. In a study by Martin et al. (Reference Martin, Doig, Heyland, Morrison and Sibbald2004) patients in intervention hospitals (nutrition intervention conducted on evidence-based guidelines) received more energy on a daily basis than those in control hospitals (5287 kJ (1265 kcal) per patient v. 4172 kJ (998 kcal) per patient) and had better outcomes (shorter hospital stay, 25 d v. 35 d, P=0·003; reduced hospital mortality, 27% v. 37%, P=0·058), although no difference in length of ITU stay was observed. However, the amount of energy received was quite low in both control and intervention patients and the authors were unable to determine the relative value of each component of the intervention. Kan et al. (Reference Kan, Chang, Sheu, Cheng, Lee and Huang2003) have defined underfeeding as <90% of requirements (defined as MEE+20%) and, in fifty-four patients who were ventilated, have found no differences in outcome (length of ITU and hospital stay, length of ventilator dependence) between those who were under-, adequately- or overfed. Mean energy intakes were found to vary from 6441 kJ (1541 kcal)/d in the underfed patients to 7812 kJ (1869 kcal)/d in the adequately-fed patients and 8715 kJ (2085 kcal)/d in the overfed patients (kJ (kcal)/kg body weight per d; 104 (24·9), 127 (30·3) and 174 (41·5) respectively). Dvir et al. (Reference Dvir, Cohen and Singer2006) have found no association between negative energy balance and length of ventilation, ITU stay, hospitalisation or mortality, although negative energy balance was shown to be correlated with ITU complications. While hypoenergetic feeding may be appropriate for critically-ill patients receiving parenteral nutrition and for critically-ill obese patients receiving enteral or parenteral nutrition, it may not be appropriate in non-obese patients. Until there are universally-agreed definitions of both under- and overfeeding, the extent to which inadequate or excess energy provision influences outcomes in patients in the ITU will remain unresolved.

The metabolic response to injury has not been specifically investigated in obese individuals, although the effects are thought to be similar to those observed in patients who are not obese. Thus, although obese patients have large fat and lean body mass stores, they are likely to develop malnutrition in response to metabolic stress, particularly if their nutritional status was poor before injury or illness. Thus, nutrition support should not be withheld. However, the determination of energy requirements is particularly problematic in obese patients (Glynn et al. Reference Glynn, Greene, Winkler and Albina1999; Horgan & Stubbs, Reference Horgan and Stubbs2003; Frankenfield et al. Reference Frankenfield, Roth-Yousy and Compher2005). Although the best predictor of BMR is body weight, there is some evidence that REE adjusted for total body weight decreases with increasing BMI in the critically ill (Zauner et al. Reference Zauner, Schneeweiss, Kneidinger, Lindner and Zauner2006). Amato et al. (Reference Amato, Keating, Quercia and Karbonic1995) have validated the formula 88 kJ (21 kcal)/kg body weight in obese patients. However, when compared with other predictive formulas for patients who are ventilated (Ireton-Jones et al. Reference Ireton-Jones1992; HBE adjusted for average weight×1·3) it was found to have a poorer predictive value (Glynn et al. Reference Glynn, Greene, Winkler and Albina1999). Whether to use actual body weight, ideal body weight or adjusted body weight remains contentious, as there is little evidence to support any particular method (Ireton-Jones, Reference Ireton-Jones2005; Krenitsky, Reference Krenitsky2005). For example, the 25% adjustment (i.e. (ABW−IBW)×0·25+IBW, where ABW is actual body weight and IBW is ideal body weight) was not developed on the basis of scientific study but was initially reported in a question-and-answer format in the Renal Dietitians Newsletter, a publication of the American Dietetic Association (Karkeck, Reference Karkeck1984; Wilkens, Reference Wilkens1984). In the light of current evidence, hypoenergetic feeding may have some advantages in critically-ill obese patients, and clinicians should avoid overfeeding metabolically-stable obese patients. Glycaemic and metabolic control and monitoring should be particularly rigorous in obese patients.

A literature review may inform the determination of energy requirements for an individual patient if there are sufficient valid data for the relevant patient group. When reviewing the literature it is necessary to take a critical approach, to be aware of the gaps in the evidence and to understand the limitations of guidelines. For many patient groups, however, there is a general lack of studies from which to make any useful recommendations. Despite the high prevalence of cerebro-vascular accident, a literature review reveals only four studies in which energy expenditure has been measured post-cerebro-vascular accident, two in ischaemic strokes (Weekes & Elia, Reference Weekes and Elia1992; Finestone et al. Reference Finestone, Greene-Finestone, Foley and Woodbury2003) and the other two in haemorrhagic strokes (Piek et al. Reference Piek, Zanke, Sprick and Bock1989; Touho et al. Reference Touho, Karasawa, Shishido, Morisako, Numazawa, Yamada, Nagai and Shibamoto1990). Another area requiring more research is the effect of abnormal losses, e.g. malabsorption disorders, wounds and fistula outputs, on energy requirements. For example, in patients with high-output fistulas the energy losses can amount to 15% of their average daily energy expenditure (Reid et al. Reference Reid and Campbell1999), while in another study that measured energy expenditure in patients with inflammatory bowel disease (some with fistulas) no account was taken of the effect of such losses (Barot et al. Reference Barot, Rombeau, Feurer and Mullen1982). Other conditions that merit more research are chronic inflammatory states, such as those that exist in patients with chronic obstructive pulmonary disease or cardiac cachexia, and the effects of different treatment modalities on the energy requirements of patients with different cancers.

The present review has shown that despite the number of studies investigating the energy requirements of different patient groups, there are still areas that require more research. In particular, the following questions need to be addressed:

• if a patient is fed to energy requirements (determined by any method) will he or she have a better outcome than if he or she is not;

• how do different levels of feeding (e.g. <BMR, BMR or BMR×stress factor, 84–105 kJ (20–25 kcal)/kg body weight) affect outcomes (e.g. mortality (in ITU or in hospital), period (d) on a ventilator, length of stay (d) in ITU, mobility and rehabilitation post ITU, length of hospital stay);

• does feeding have to be tailored to an individual's estimated requirements or would a limited number of ‘standard’ regimens be sufficient, e.g. 4180, 6270 or 8360 kJ (1000, 1500 or 2000 kcal)/d;

• do prescriptions for energy need to be altered to reflect MEE on a day-to-day basis? Will this approach result in a better outcome when compared with a stable energy prescription over the same period;

• what clinical variables should be used to assign specific stress factors and which should be used to monitor change? How often should variables be monitored;

• what are the energy requirements of patients receiving long-term nutritional support at home or in the community.

It is likely that recommendations for determining energy requirements will be different for obese and undernourished individuals and for patients at risk of re-feeding, when compared with well-nourished individuals. Similarly, different metabolic states (whether or not patients are hospitalised) are also likely to have a profound effect on requirements.

In clinical practice, therefore, it is important to recognise that estimated energy requirements are only a starting point, and that all clinicians should regularly review their patients to ensure they are meeting their nutritional goals and to evaluate the effectiveness of nutritional support. Energy requirements may change as a result of changes in a patient's clinical condition, nutritional status and prognosis, and to date there appear to be few studies that measure the changes in energy expenditure that occur as a patient's condition changes. Furthermore, the likely duration of nutrition support needs to be considered. In the short term, a period of slight over- or underfeeding may not be problematic in metabolically-stable individuals, although it is unlikely to be the case in the critically ill. Many patients receive long-term feeding at home or in other community settings, and even in the metabolically-stable individual prolonged over- or underfeeding is likely to have adverse clinical effects, especially in the absence of adequate monitoring and follow-up.