Effect of nutritional support on clinical outcome in patients at nutritional risk

ARTICLE IN PRESS Clinical Nutrition (2004) 23, 539–550

www.elsevier.com/locate/clnu

ORIGINAL ARTICLE

Effect of nutritional support on clinical outcome in patients at nutritional risk Niels Johansena,*, Jens Kondrupa, Lise Munk Pluma, Line Baka, Pernille Nørregaarda, Else Bunchb, Hanne Bærnthsenb, Jens Rikardt Andersenc, Irene Højlund Larsenc, Anette Martinsenc a

Clinical Nutrition Unit 5711, Rigshospitalet, University Hospital of Copenhagen, Blegdamsvej 9, DK-2100 Copenhagen, Denmark b Department of Surgery, Hobro-Terndrup Hospital, Denmark c Department of Medicine, County Hospital, Nykbing Falster, Denmark Received 3 July 2003; accepted 7 October 2003

KEYWORDS Undernutrition; Hospital; Intervention; Randomised controlled trial; Length of stay

Summary Background & Aims: Undernourished patients have an increased risk of complications and a prolonged hospital stay, compared to those who are not undernourished. The aim of this study was to evaluate the effect of nutritional intervention in a random sample of hospitalized patients at nutritional risk. Methods: A randomized, controlled trial of nutritional intervention in 212 patients. Intervention consisted of a specialized nutritional team (nurse and dietician) who attended patients and staff for motivation, detailed a nutritional plan, assured delivery of prescribed food and gave advice on enteral or parenteral nutrition when appropriate. The control group received the standard regime used in the department. The primary endpoint was the part of the length of stay (LOS) that was considered to be sensitive to nutritional support, designated LOSNDI. The nutritional discharge index (NDI) consists of three objective criteria: (1) the patient is able to manage toilet visits without assistance, reflecting mobilization; (2) the patient is without fever (tpo381C), reflecting absence of infection; and (3) the patient has no intravenous access, reflecting absence of complications in general. On the day when all three criteria were fulfilled, hospital stay was no longer considered to be sensitive to nutritional support. Actual LOS is also reported. Incidence and severity of complications were recorded to explain LOSNDI findings. As a secondary endpoint, quality of life was evaluated by the Short Form 36 (SF-36) questionnaire. Results: Intervention led to an intake of X75% of requirements in 62% of the intervention patients, as compared to 36% of the control patients. Rates of complications, mean LOSNDI and LOS were not significantly different between the two study groups. However, among patients with complications a difference in LOSNDI between intervention patients (1472 days, mean7SE) and control patients (2072 days) was statistically significant (P ¼ 0.015). In the same patients, LOS was 1772 days in the intervention group and 2272 days in the control group (P ¼ 0.028). The SF-36 questionnaire did not show a significant effect of treatment.

*Corresponding author. Tel.: þ 45-35-45-23-59; fax: þ 45-35-34-29-13. E-mail address: [email protected] (N. Johansen). S0261-5614/$ - see front matter & 2003 Elsevier Ltd. All rights reserved. doi:10.1016/j.clnu.2003.10.008

ARTICLE IN PRESS 540

N. Johansen et al.

Conclusions: Protein and energy intake of nutritionally at-risk patients was increased which resulted in shortening of the part of the length of stay that was considered to be sensitive to nutritional support (LOSNDI) and shorter length of stay (LOS) among patients with complications. & 2003 Elsevier Ltd. All rights reserved.

Introduction A large proportion of hospital patients is undernourished.1–4 McWhirter and Pennington found that 40% of the patients were undernourished on admission to hospital, and that 75% of the undernourished patients who remained in hospital for more than 1 week lost further weight.1 Undernutrition is, in particular, associated with loss of muscle strength and impaired immune function,5,6 which lead to an increase in complications and a prolonged hospital stay.7 A recent meta-analysis suggests that nutritional support reduces complication rates, infection rates, mortality and length of hospital stay and improves quality of life.8 However, these studies were nearly all conducted in specific groups of patients (e.g. gastro-surgery, chronic liver disease, chronic pulmonary disease, cancer patients, elderly) and the generalizability of these results can be discussed. A new screening method for general use in hospitals (NRS-2002, adopted by ESPEN) is based on an analysis of such controlled trials.9,10 In the present study, the aim was to evaluate the clinical benefit of nutritional intervention in a random sample of all patients at nutritional risk according to NRS-2002 from three different hospital levels. The investigation was carried out as a prospective, randomized, controlled trial. To allow for the expected heterogeneity of the patients, we chose length of stay, and particularly the length of stay considered to be sensitive to nutritional support, as primary endpoints. Incidence and severity of complications were used as explanatory variables. As a secondary endpoint, quality of life was investigated. Further, the interobserver variation of the NRS-2002 screening method was evaluated.

Materials and methods Patient accrual and intervention The study was in accordance with the Helsinki declaration II and acknowledged by the ethical committee of Copenhagen (KF 01-025/01). The

following hospitals participated: Rigshospitalet University hospital Copenhagen (RH), Nykbing Falster County hospital (NF), Hobro-Terndrup local hospital (HS). At each hospital, a nurse and a clinical dietitian formed a team. Every weekday, the team selected a minimum of five patients among patients admitted during the preceding 24 h, by a random numbers system. The following patients were excluded: less than 4 days’ expected admissions, less than 18 years of age, less than 1 month of expected survival, patients who did not understand the Danish language, previously participating patients, patients who were placed next to another participant in the same room, pregnant or lactating healthy women, patients with psychiatric disorders, patients in haemodialysis and patients who were already receiving, or were planned to receive, a standard parenteral or PEG-tube feeding. It was required per protocol that patients were randomized and entered the study within 36 h after admission. The team screened the selected patients according to the NRS-2002 system which is based on an analysis of previous randomized controlled trials of nutritional intervention.9 Nutritional risk is evaluated by two components: nutritional status and severity of disease. Nutritional status is evaluated by three variables: body mass index (BMI), recent weight loss and dietary intake during the last week before admission. A score of 3 is given for severe undernutrition: BMIo18.5, weight loss 45% during the last month, or an intake of 0–25% of requirement. A score of 2 is given for moderate undernutrition: 18.5o BMIo20.5, weight loss 45% during the last 2 months, or an intake of 25–50% of requirement. A score of 1 is given for mild undernutrition: weight loss 45% during the last 3 months, or an intake of 50–75% of requirement. For severity of disease, as an indicator of stress metabolism and increased nutritional requirements, a score of 3 is given for severe illness, e.g. intensive care or sepsis, a score of 2 is given for moderate illness, e.g. colectomy, and a score of 1 is given for mild illness, e.g. fractured neck of femur. Finally, the score for nutritional status is added to the score for severity of disease to give a total score, which can range from 0 to 6. Patients with a

ARTICLE IN PRESS Effect of intervention in patients at nutrional risk

total scoreX3, i.e. defined as being nutritionally at risk, were offered to participate in the study. Patients who were expected to achieve a scoreX3 due to a scheduled major operation, or a similar stress-inducing treatment, were also offered to participate. Patients with a scoreo3 were reevaluated once a week, and in case that a score X3 developed during the admission, they were also offered to participate. After informed consent, the patients were stratified by age: oor X60 years, and randomized in blocks of 6, using sequentially numbered sealed non-transparent envelopes, which were opened to allocate patients into one of two groups: control or intervention. The procedure was repeated until five patients in each hospital were active in the study at any point of time. Patients in the control group received nutritional care according to the usual procedure of the department. Intervention patients in addition received daily attention from the team consisting of: (1) motivation of patient and staff, (2) adjusting the nutritional plan by estimation of protein- and energy requirements and ordering food in collaboration with the patient and (3) securing the supply of food ordered. In addition, the team advised patients and staff about when to change to tube feeding or parenteral nutrition. The team also estimated dietary intake by 24-h dietary recording by staff and/or patient or, if not available, by 24-h dietary recall. In addition, the team measured body weight three times a week. Energy requirement was estimated by the factorial method, based on BMR calculated by the Harris–Benedict equation 11 and protein requirement was taken to be 18% of energy requirement, according to official Danish guideline.12

Outcome variables Length of stay Previous clinical trials have used different outcome variables, as seen to be relevant to specific patient groups.8,9 The most common outcome variables used were infections in studies of medical or surgical patients, and wound problems or all complications in studies of surgical patients. However, some general outcome variables were also found to be sensitive to nutritional intervention, e.g. muscle function, mobilization or quality of life. In some studies, these effects led to reduced length of stay or decreased mortality. Because our study population would be heterogeneous in terms of admission diagnoses and therefore also in outcome variables relevant for the diagnoses, we

541

considered length of stay (LOS) to be a suitable common denominator. However, LOS is dependent on many factors, some of which are modified by nutritional support, e.g. mobilization and infections, and some of which are not, e.g. planned admission for investigative purposes, delay until initiation of homecare or patients’ preferences. Therefore, we created a measure for the part of the LOS which is considered to be sensitive to nutritional support, inspired by a study in surgical patients.13 When the patient fulfils three objective criteria: (1) is able to manage toilet visits without assistance, reflecting mobilization, (2) is without fever (tpo381), reflecting absence of infection, and (3) is without intravenous access, reflecting absence of complications in general,8,9 hospital stay is no longer considered to be sensitive to nutritional support, and the patient is defined as being ready for discharge in a nutritional sense. This nutritional discharge index (NDI) is followed daily, and the day on which all three criteria are fulfilled (NDI ¼ 3), and maintained for the rest of the admission, is designated LOSNDI. Since our study had a scheduled time limit, we decided to follow patients for a maximum of 28 days after inclusion. In a previous study,11 90% of the at-risk patients were discharged within 28 days. The time from admission until discharge (with a maximum of 28 days of study inclusion) was designated: LOS28. To summarize: LOS28 ¼ LOS from admission until inclusion þ LOS from inclusion until discharge (max. 28 days). LOSNDI ¼ LOS28number of final days with NDI ¼ 3. To explain differences in LOSNDI and LOS28, we evaluated the incidence and severity of complications and, as a secondary outcome measure, we evaluated changes in quality of life. Both incidence and severity of complications have been shown to be sensitive to nutritional intervention8,9,14 and the Short Form 36 (SF-36) quality of life (QoL) questionnaire has been found to be sensitive to nutritional support.15 Complications Complications, i.e. unexpected events during admission, were graded as minor or major according to the study of surgical patients by Buzby et al.16 In their study, the grade of complications (minor or major) was based on previous experience with LOS associated with various complications, indicative of the severity of a complication except for death. In the present study, the system was modified to include non-surgical patients. Infectious complications not mentioned in Buzby et al.16 were defined according to the US Centre of Disease Control17 and

ARTICLE IN PRESS 542

graded as minor or major in similarity to Buzby et al.,16 by judgement of the authors prior to opening the randomization code. Likewise, noninfectious complications were discussed and graded as minor or major by the authors prior to opening of the randomization code. If a patient experienced more than one complication, only the most severe was counted. A member of the investigation team blinded to patient allocation (NJ), phoned the wards daily to collect the following data from the nurses responsible for the patients’ care: clinical events to define the rate of complications, temperature, presence of intravenous access, the need for assisted toilet visits. In the case of fever (tpX381C), data on site of infection, bacteriology and use of antibiotics were collected. The nurses were instructed by the team not to reveal group allocation of the patients to NJ. Quality of life Quality of life was evaluated by the Danish version of the SF-36 QoL questionnaire that takes about 5 min to answer.18 The SF-36 QoL focuses on health related to social function and consists of 36 questions within eight dimensions covering various physical and mental aspects: physical functioning, role limitations due to physical problems, bodily pain, general health, vitality, social functioning, role limitations due to emotional problems, mental health. The answers to the questions within each dimension are converted to a score from 0 to 100 and these results are weighted and summarized in a physical component score and a mental component score on a scale from 0 to 80.19 The team gave the patients a SF-36 QoL questionnaire to complete at inclusion and on day 28 if the patient was still present at the hospital. If the patient was not able to read the questionnaire or to write the answers, the patient was interviewed by the team, who then completed the questionnaire in the patient’s presence. Patients discharged before day 28 received the follow-up questionnaire by mail on day 28. If it was not returned, NJ sought to administer the questionnaire as a telephone interview. Interobserver variation While performing the intervention study, the three members of the RH investigating team: a nurse, a dietitian and an MD (NJ) evaluated the interobserver agreement of the screening tool. To reach a conclusive result, the interrater agreement corrected for random agreement (i.e. a k-value 40.60 for an acceptable agreement) had to be examined in X40 patients.20 Patients were selected consecutively among those from the main study. Each investigator screened the patients within 24 h of

N. Johansen et al.

admission. If not present in the ward, they were visited on the following day (within 48 h of admission). If patients were dismissed from the hospital before all three investigators had performed their screening, they were excluded from the study. Each investigator was unaware of the results obtained by the other investigators and patients were not discussed until all participants had been screened. Patients were categorized as being nutritionally at risk or not at risk. Results were expressed as observed frequencies of agreement between investigators, and adjusted for agreement by chance by calculating a k-value.20 Altogether 63 patients were selected; two patients were discharged before screening by all three investigators and 61 patients were evaluated: 31 males and 30 females.

Statistical analysis Given a specified time limit, we expected to include 200 patients in the study. Unpublished data from a literature analysis9 suggested a 20% shortening of LOS by nutritional intervention; with a significance level a ¼ 0.05, the estimated testpower was ¼ 0.45. This rather low test-power could lead to a falsely negative result, which would lead to a false rejection of the screening tool and/or the efficacy of our intervention. It would not, however, affect the perception of clinical efficacy of nutritional intervention, as documented in the metaanalysis of nearly 70 RCTs.8 Analyses were performed on an intention-to-treat basis or in subgroups as specified. Results are presented as mean 7SEM. Statistical difference between subgroups was tested by ANOVA for continuous data. When interaction was present (Table 5), the results were further analysed by including the interaction as a variable in a generalized linear model. Categorical data were analysed using Fisher’s exact test. Wilcoxon’s test was used for a one sample test. The following software was used: GraphPad version 3.02, GraphPad Software, San Diego, California, USA; Systat 10, SPSS Inc. Chicago, II, USA, and for confirmation, SAS package software version 8.2, SAS Institute Inc., Cary, NC, USA.

Results Interrater agreement Among the 61 patients, the three investigators found 18,16 and 23 patients to be nutritionally at risk. Agreement on nutritional risk (grouping

ARTICLE IN PRESS Effect of intervention in patients at nutrional risk

patients as ‘‘at risk’’ or ‘‘not at risk’’) between all three investigators was found to be 77%, with a kvalue ¼ 0.67.

Patients From August 1st 2001 to March 1st 2002, a total of 35.442 patients were admitted to the three hospitals. By the random numbers system, 7.468 patients were selected. Of these, 2.436 were excluded due to a o4 day’s expected stay (same day surgery, etc.), 1.292 were children or pregnant women, 434 were in haemodialysis, intensive care and/or had a fixed artificial feeding schedule, or had mental problems, or were terminally ill; 81 patients were undergoing operation at the time of the first visit, and were judged not to be able to be included with consent and with surgeons’ permission to eat within 36 h after admission, as required per protocol. Of the remaining patients, 2.915 were not nutritionally at risk, leaving 310 patients eligible for entering the study. Among these patients, informed consent could not be obtained within 36 h in 14 patients, since a minimum of 10 h was required for consideration by the patient. In addition, 81 patients directly refused to participate, in most cases because they were reluctant to engage in more activities early after their admission (N ¼ 50), for unknown reasons (N ¼ 25), or because they were depressed by their clinical condition (N ¼ 6). Of the 215 patients randomized, two patients withdrew from the study after 4 and 11 days due to terminal illness and one patient turned out to have been falsely categorized as nutritionally at risk. These three patients were removed from the data set before opening the randomization code. The intention-to-treat sample thus consisted of 212 patients (RH:67, NF:82, HS:63): 104 patients in the Table 1

543

control group, 108 patients in the intervention group. Patients’ characteristics are shown in Table 1. The two groups were similar with respect to age, sex, and nutritional status at admission. They were moderately to severely undernourished and, on average, had a mild grade of severity of disease. The distribution of diagnoses at admission (Table 2) were similar between the two groups.

Intake Data on intake from five intervention patients and five control patients were not obtained due to a very short admission time (1–2 days). In 23 of the 103 intervention patients and in 31 of the 99 control patients, data on protein and energy intake were collected by dietary recall only, but in all other patients by dietary recording. In the intervention group, 85 patients were fed solely by food and/or liquid supplements, compared to 86 in the control group. In the intervention group, 18 patients were fed artificially part of the time (10 patients were tube fed for 40710% (mean7SE) of the study-time, 11 patients were parenterally fed for 4579% of 5 the study-time, and oral intake was maintained for 8973% of the study-time). In the control group, 12 patients were fed artificially part of the time (six patients were tube fed for 52714% of the study-time, seven patients were parenterally fed for 4576% of the study-time, and oral intake was maintained for 8375% of the study-time). Table 3 shows that intervention patients achieved a statistically significant higher mean intake of energy and protein, both when expressed in absolute amounts and in percentage of calculated requirements. However, a difference in mean values can obscure a significant proportion of severely underfed patients in either group. This

Base-line characteristics. Intervention

Control

N Age (years, mean7SE) Sex (M/F)

108 62.071.6 54/54

104 62.471.7 48/56

Intake before admission (quartiles of estimated requirements): 0–25% or 25–50% 50–75% or 75–100%

52 56

54 50

Body weight kg (mean7SE) Body mass index (kg/m2) (mean7SE) Score for nutritional status (0–3) (mean7SE) Score for severity of disease (0–3) (mean7SE)

61.571.65 21.270.50 2.470.08 1.070.07

62.471.51 21.870.48 2.570.07 1.070.06

ARTICLE IN PRESS 544

Table 2

N. Johansen et al.

Diagnosis at admission.

Diagnosis Medical departments: Cardio-vascular disease: Ischaemic heart disease Incompensated heart disease Other cardio-vascular Respiratory disease: Chronic obstructive pulmonary disease Pneumonia Other acute respiratory illness Neurological disorders: Vascular disease Other neurological illness Gastrointestinal disorders: Upper G.I.-disorder Lower G.I.-disorder Pancreatic disease, hepatic cirrhosis Various internal medicine: Infection Diabetes mellitus Observation Nephrology Malignant disease (oncology/haematology) Surgical departments: Neurological surgery Chest surgery Major abdominal surgery Minor abd. Surgery, incl. appendicitis Abdominal pain Major orthopaedic surgery Fractured neck of femur Trauma Other surgery Malignant disease (oncology/haematology)

may be important, since the efficacy of nutritional intervention may depend on a threshold intake, rather than on the mean intake of a group, in particular if intervention leads to a higher intake only in patients who have a borderline insufficient intake. Severe underfeeding was defined as an intake o75% of estimated requirements, since all patients with an intake lower than this lost weight in a previous study.21 Fewer patients were severely underfed in the intervention group, as compared to the control group (Table 3). When a patient was severely underfed, the teams recorded the reason for treatment failure. After 4 days of intervention, the teams no longer gave insufficient time to reach dietary goals as a reason. For these patients (44 days of inclusion), 70% of the intervention patients (54 of 77) reached X75% of estimated requirements, as compared to 41% of control patients (30 of 74). Causes of intake o75%

Intervention

Control

2 1 4

1 3

7 6 1

8 8 3

2 4

4 2

6 5 2

1 8 5

3 3 17 2 9

5

1 1 4 7 3 1 1 5 11

11 11 1 1 3 1 11 1

7 9

of requirement among the 23 treatment failure in the intervention group were: development of a terminal condition (eight patients), lack of motivation (patient or staff, eight patients), fasting for procedures (two patients) and unknown five patients). When disregarding terminally ill patients and patients fasting for procedures, the minimum intake was reached in 81% of the intervention patients after 4 days of intervention (Table 3). Two or more weight measurements were obtained only in 57 intervention patients and 52 control patients. In most cases, the missing weighing was due to the patient being immobile. Weight changes in patients without oedema revealed no difference; intervention group: 0.2270.54 kg (mean7SE, N ¼ 53), control group: 0.1070.31 kg (N ¼ 42). The mean time between the two measurements was 1071 days.

ARTICLE IN PRESS Effect of intervention in patients at nutrional risk

Table 3 Intake of energy and protein during the study period.

Number of patients Intake (mean7SE) Energy (kJ/kg per day) Protein (g/kg per day) Energy (% of requirement) Protein (% of requirement) % of all patients with protein intake X75% of requirement % of selected patients with protein intake X75% of requirementz

Intervention

Control

103

99

n

12675

10675

1.170.04

0.970.04n

9973

8473n

8373

6673n

62

36w

81

46y

n

Po0.005 (Student’s t-test). P ¼ 0.0004 (Fisher’s exact test). z Patients with 44-day intervention who were not terminally ill and who were not forced to starvation due to procedures or surgery (N for intervention and control groups ¼ 67 and 65, respectively; see text for details). y Po0.0001 (Fisher’s exact test). w

Complications and antibiotics There was no statistically significant difference between the two study groups, neither in total, minor or major complications (Table 4). In the control group, LOS28 was 2172 days (mean7SE) for those with a major complication and 1472 days for those with a minor complication (P ¼ 0.01), supporting the grading of minor and major complications based on LOS. The difference in LOS28 between the groups will be dealt with below. As analysed by logistic regression, complications were related to severity of disease (OR for each score unit: 2.5 with a 95% CI: 1.3–4.6; P ¼ 0.005), but not to other base-line characteristics (age, sex, coding for hospital, score for nutritional status, medical versus surgical admission), nor to performance of surgery. Antibiotics were given orally or intravenously to 52 intervention patients and 60 control patients for an average of 3.970.6 days (mean7SE) in the intervention group, and 4.070.5 days in the control group. For patients receiving antibiotics due to an infectious complica-

545

Table 4 Incidence of complications. Severity of complication

Minorn Majorw Total

Number of patients with complications Intervention

Control

10 24 34

6 17 23

n

Urinary tract infection (intervention: 7, control: 4), noninfectious diarrhoea (intervention: 2, control: 2), wound infection (intervention: 1). w Death (intervention: 9, control: 6), pneumonia (intervention: 4, control: 4), septicaemia/bacteraemia (intervention: 4, control: 2), empyema (intervention: 2), gastroenteritis (control: 1), cardiac, renal or hepatic failure (intervention: 2, control: 2), cerebral bleeding (intervention: 1), gastrointestinal obstruction (intervention: 1, control: 2), thrombosis (intervention: 1).

tion, i.e. a nosocomial infection (20 intervention patients and 12 control patients), the averages were 10.171.5 days and 10.271.6 days, respectively.

Length of stay (LOSNDI and LOS28) Table 5 shows LOSNDI in the two groups. Patients who died (intervention group: 9, control group: 6) were excluded from this analysis, because death by itself shortens LOS. An additional difficulty arose from the finding, that some of the patients had a maximum NDI at the time of inclusion and maintained it during the entire study period. These patients by definition had a LOSNDI ¼ 0 and nutritional intervention could not be expected to improve this LOSNDI. Therefore, in the primary analysis (Table 5), we also excluded these patients (17 intervention patients and eight control patients). At the end, these patients will again be included in the complete analysis. For all patients in Table 5, there was no difference in LOSNDI between the groups. A subgroup analysis was undertaken to examine whether the intervention had an effect on the severity of complications, i.e. LOSNDI among patients with a complication. Operation by itself also prolonged LOSNDI and was therefore included as a separate grouping variable. It appears from Table 5, that LOSNDI among patients with complications and no operation was shorter in the intervention group than in the control group. An analysis by ANOVA indicated interaction between complication and group (control/intervention), i.e. that the increase in LOSNDI associated

ARTICLE IN PRESS 546

N. Johansen et al.

with complications was less in the intervention group. Therefore, the analysis was extended in a generalized linear model with the interaction term (group  complication) as a variable in addition to the variables mentioned above (group, complication, operation). This resulted in the following P-valuesFgroup: P ¼ 0.026; complication: Po0.0005; operation: P ¼ 0.002; group  complication: P ¼ 0.011. With this model, the estimated effect of intervention in patients with complications was 5.972.2 days (mean7SE, P ¼ 0.015 by pairwise comparison with a Bonferoni correction for multiple comparisons), agreeing closely with the difference observed in Table 5. It was further examined whether other variables contributed to this effect by adding base-line characteristics (age, sex, degree of undernutrition, severity of disease, code for hospital, code for medical or surgical admission) to the generalized model, but this was not the case. The lack of a difference in LOSNDI for all patients in Table 5 is

related to the fact, that a total of 32 of the intervention patients experienced a complication and/or operation versus 21 of the control patients (P ¼ 0.032). Due to the effect of intervention in patients with complications, this uneven distribution did not prolong the mean LOSNDI for all intervention patients. When adjusting for the uneven distribution of patients in the subgroups (by weighted averages of means and variances), the calculated LOSNDI in the intervention group was 9.270.7 days and 10.270.7 days in the control group (NS). Table 6 shows the analysis extended to actual LOS (LOS28) and the intention-to-treat sample. In the primary analysis, the estimated effect of intervention on LOS28 was 5.872.3 days among patients with complications (17.271.6 versus 22.271.8). With adjustment for the uneven distribution of patients in the subgroups, the calculated LOS28 in the intervention group was 11.870.7 days and 12.770.7 days in the control group (NS).

Table 5 Length of stay until the patient no longer needed assistance in toilet visits, had no fever and had no i.v. access (LOSNDI). LOSNDI Intervention

All Subgroup No complication, no operation Complication, no operation No complication, operation Complication, operation

Control

N

Mean7SE

N

Mean7SE

82

9.970.8

90

9.270.8

50 18 9 5

7.170.9 14.071.5n 12.472.1 17.472.9

69 14 7 0

6.970.8 19.671.7 11.672.4

In this primary analysis, patients who died and patients with LOSNDI ¼ 0 were excluded. See text for details. n Compared to control group with compliations: P ¼ 0.015 (ANOVA, see text for details).

Table 6 Length of stay until the patient no longer needed assistance in toilet visits, had no fever and had no i.v. access (LOSNDI) or actual length of stay until 28 days (LOS28) among patients from the primary analysis (as in Table 5) and the intention-to-treat sample (ITT). LOS28

LOSNDI

All Primary ITT Complication, no operation Primary ITT n

Intervention

Control

Intervention

Control

9.970.8 (82) 8.270.8 (99)

9.270.8 (90) 8.570.8 (98)

12.570.9 (82) 11.670.8 (99)

11.770.9 (90) 11.570.8 (98)

14.071.5 (18)n 12.071.5 (21)z

19.671.7 (14) 17.171.7 (16)

17.271.6 (18)w 17.071.5 (21)

22.271.8 (14) 20.271.7 (16)

Compared to control group with complications: P ¼ 0.015. Compared to control group with complications: P ¼ 0.028. z Compared to control group with complications: P ¼ 0.031 (ANOVA, see text for details). w

ARTICLE IN PRESS Effect of intervention in patients at nutrional risk

In the intention-to-treat analysis (though still excluding those who died), the estimated effect of intervention on LOSNDI was 5.372.2 days among patients with complications (12.071.5 versus 17.171.7). With adjustment for the uneven distribution of patients in the subgroups, the calculated LOSNDI was 7.670.7 days in the intervention group and 9.270.7 days in the control group (P ¼ 0.11). The estimated effect of intervention on LOS28 was 4.072.2 days (P ¼ 0.15; 17.071.5 versus 20.271.7). With adjustment for the uneven distribution of patients in the subgroups, the calculated LOS28 was 11.270.7 days in the intervention group and 12.270.7 days in the control group (NS).

Components of the nutritional discharge index The association between LOSNDI as the dependent variable and the days on which the individual NDIcriteria were fulfilled as independent variables was examined by a multiple linear regression analysis. All patients, except those who died, were included: LOSNDI ¼ 1.31 (P ¼ 0.05) þ 0.85  i.v.-access (Po 0.005) þ 0.40  Ass. toilet visit (Po0.005) þ 0.04  Fever (P ¼ 0.3). Within the group of patients with complications described above (Table 5), an analysis of the three criteria of LOSNDI by the generalized model showed the intervention to reduce the number of days with i.v. access (4.9day reduction; P ¼ 0.015), but the two other components were not significantly reduced. To evaluate the relationship between LOS28 and LOSNDI, a correlation analysis was performed, also in all patients except those who died (Fig. 1): LOSNDI ¼ 0.90  LOS281.12 (r2 ¼ 0.85, Po0.0001). The intercept on the x-axis was 1.24 days. However, this contrasts with the 3-day difference in means of LOSNDI and LOS28 given above. Therefore, a Bland–Altman analysis 22 was carried out, giving a bias and a limit of agreement of 2.475.8 days (mean72SD), indicating that LOSNDI and LOS28 had a rather limited agreement. It is clear from Fig. 1

Table 7

547

that over the whole spectrum of LOS28, many patients had a relatively much lower LOSNDI.

SF-36 QoL questionnaire Of 212 patients, 170 returned the first questionnaire (80%): 21 patients were unable to return the questionnaire due to unconsciousness or other serious conditions and 21 patients withdrew from this part of the study. Of the 170 patients, the second questionnaire was returned by 122 patients, i.e. 57% of the original 212 patients. Sixteen were unable to return it because of serious conditions, 19 died and 13 withdrew. A drop-out analysis showed that respondents to both questionnaires were comparable to non-respondents, according to the base-line characteristics in Table 1. Also, there were no statistically significant differences in baseline characteristics between intervention and control patients in the group of respondents. In Table 7, the two summary scores are shown. There was no statistically significant improvement

Figure 1 Correlation between observed length of stay (LOS28) and length of stay until the patient no longer needed assistance in toilet visits, had no fever and had no i.v. access (LOSNDI). Correlation between LOS28 and LOSNDI among the individual patients. Patients with LOSNDI ¼ 0 were excluded from the correlation analysis, but are included in the figure (see text for explanation). The solid line represents the regression line. Results from a Bland–Altman analysis is given in the text.

Quality of life measurement for physical and mental summary scores (mean7SE). Physical score

N Initial assessment (A) Final assessment (B) Change (BA)

Mental score

Intervention

Control

Intervention

Control

52 31.571.2 33.371.4 2.471.3

58 33.171.3 33.771.3 0.271.5

52 40.171.9 41.371.9 2.272.5

58 39.071.8 42.071.8 3.372.0

ARTICLE IN PRESS 548

from base-line in any of the study groups. However, there were improvements of individual health items. Observed changes tested versus zero changes by a Wilcoxon one sample test, showed improvements in bodily pain ¼ 10.375.0 (mean7SE, P ¼ 0.04), social functioning ¼ 10.575.2 (P ¼ 0.05) and mental health ¼ 7.973.2 (P ¼ 0.02) in the control group, and, bodily pain ¼ 10.074.8 (P ¼ 0.04) and social functioning ¼ 15.576.0 (P ¼ 0.01) in the intervention group. An additional analysis similar to the one performed for LOSNDI in Table 5, both with delta values and final values adjusted for initial values (ANCOVA analysis), did not either reveal any effect of nutritional intervention on any item. A separate analysis of respondents with complications (13 intervention patients and eight control patients) showed no differences. When analysing single questions related to physical capability (reflecting mobilization), the intervention group showed a statistically significant improvement in walking 100 m (P ¼ 0.047), whereas no improvement was seen in the control group.

Discussion The NRS-2002 screening system9 is based on an analysis of previous intervention trials by relating reported clinical outcome to scores for nutritional status and severity of disease in each study. The present study confirms that patients identified by this system will improve clinically from nutritional intervention, since the severity of complications was reduced as shown by reduced LOSNDI (Table 5) and LOS28. This effect in patients with complications can be ascribed to nutritional intervention: initial intakes of energy and protein in intervention and control patients with complications were similar to the data for all patients shown in Table 3, but the intervention group maintained its higher intake during the period with complication while the control group had a considerable further decrease when the complication occurred (data not shown). In addition, differences in base-line characteristics did not contribute to the effect on LOSNDI (see Results). The uninterrupted higher intake in the intervention group during periods with complications underlines the role of the treatment team in assuring continuity of care and immediate reactions to failing intake. In contrast to the original hypothesis of a 20% reduction in LOS, the observed LOSNDI and LOS for the whole group of intervention patients were not reduced (Table 5). This was in part due to the fact

N. Johansen et al.

that the intervention group had a significantly higher number of patients with complications and/or operations which would prolong LOSNDI and LOS. The uneven distribution occurred despite the similarity of patients’ base-line characteristics, and it is probably caused by the limited number of patients in some of the subgroups (Table 5). The best estimate of the overall effect on LOS is given by the adjusted intention-to-treat analysis (see Results) which suggested a (statistically insignificant) 1-day reduction in LOS, equivalent to an 8% reduction, mainly a result of the nearly 20% reduction among the patients with complications. It was also expected that the rate of complications would be lower in the intervention group, in agreement with the recent meta-analysis.8 However, most studies showing a reduction in the occurrence of complications have dealt with surgical patients,9 and in our study only 21 out of 212 patients were subjected to surgery. The exclusion of 81 patients being operated upon during the first day of admission (when the nutritional risk screening was meant to take place) has probably left out a number of patients who were nutritionally at risk and who might have had complications that could be prevented by nutritional support. Nutritional intervention had no effect on LOSNDI and LOS28 in the majority of patients who did not have complications. One reason could be an insufficient difference in intake. In an analysis of previous trials,9 the mean intakes of energy and protein were 8478 kJ/kg per day and 0.770.1 g/kg per day in the control groups and 15578 kJ/kg per day and 1.470.1 g/kg per day in the intervention groups, equal to a doubling in mean intake in the intervention groups. In the present study, the mean intakes of the control and intervention groups were higher and lower, respectively, than in the previous trials, and the difference in mean intake was only about 20%. It may have been implicit in the previous trials to include patients with an insufficient intake and since most studies were designed to test artificial feeding, the intakes were higher. It should be noted, however, that despite the modest difference in mean intake in our study, only onethird of the control patients reached a minimum of 75% of estimated requirements whereas this was the case for two-thirds of the intervention patients (Table 3). After 4 days of intervention, the minimum goal was reached in 81% of the intervention patients when disregarding patients who became terminally ill or were forced to starvation because of procedures, etc. Still, a difference in outcome might have occurred in patients without complications, if sufficient intake after hospital admission had been an exclusion criterion, and/or

ARTICLE IN PRESS Effect of intervention in patients at nutrional risk

if more artificial feeding had been used to reach at least the minimum requirements in all patients. The lack of a more general effect could also be due to the fact that only 30 patients (14 intervention and 16 control patients) included in the present study actually were similar to those included in previous trials: chronic obstructive pulmonary disease (N ¼ 15), cirrhosis of the liver (N ¼ 7), fractured neck of femur (N ¼ 1), major abdominal surgery (N ¼ 7), i.e. that the effect seen in previous diagnosis-specific RCTs cannot be reproduced in a general hospital population due to all the other variables determining clinical outcome. Finally, the patients not having complications had a rather short LOS (9.470.8 days in the control group) and in a recent meta-analysis,8 it can be seen that in studies with a LOSo10–12 days in the control groups, no studies showed a reduction in LOS. In summary, the screening method may be too unspecific to include only patients who will benefit from nutritional support, since it includes patients who have a reasonable spontaneous dietary intake, it includes patients with a relatively short length of stay and it includes categories of patients in whom a clinical benefit is not documented. However, the patients with complications, in whom an effect on LOSNDI and LOS28 was demonstrated, also included patients who initially had a reasonable spontaneous dietary intake and who were not similar to those of previous RCTs. It will not be easy to predict which at-risk patients who will later develop a complication and/or an insufficient dietary intake, and a delay of intervention until insufficient intake over several days has been documented, or a complication occurs, will probably lessen the clinical effect of intervention. Therefore, the treatment of all patients at risk seems necessary to obtain the clinical benefit in those who later develop complications. It can also be discussed whether the NDI proved to be a suitable measure of the clinical effect. It is obvious that LOS28 and LOSNDI do not carry the same information (Fig. 1): one-third of the patients had a LOSNDI that was less than two-thirds of the LOS28, and the Bland–Altman analysis showed a considerable 95% CI of the mean difference of 2.4 days. The analysis of the association between the three NDIcriteria and LOSNDI showed LOSNDI to be related to assisted toilet visits and i.v. access, but not to fever. This suggests that fever could have been omitted as a component of NDI. In most infections, antibiotics were administered intravenously at least until body temperature had normalized. Therefore, most days with fever were included in i.v. days which could explain why i.v. access, and not fever, was related to LOSNDI. In a different

549

group of patients, with infections treated more by the oral route, it could still be relevant to include fever as a variable, since it is unlikely that patients are considered ready for discharge with a temperature X381. Another problem was the approximately 10% of the patients who had a maximum NDI at admission and maintained it for the entire hospital stay, since these patients could not be improved by intervention. It is, however, not reasonable to exclude such patients from the study a priori, since a large part of the patients who have a maximum NDI at admission will become submaximal during the admission, e.g. due to a complication or an operation. LOS is an inaccurate outcome variable for many reasons and will always need a further exploration before being accepted as a result of intervention (see Materials and methods). LOSNDI has the specificity of information lacking in LOS and, in addition, the analyses given in the results section (Table 6) showed that the Pvalues for differences in LOSNDI in all cases were less than the corresponding P-values for LOS28, indicating that LOSNDI was more precise. Finally, in a multicentre trial, local policy may affect LOS whereas LOSNDI is easier to standardize. For these reasons, we conclude that LOSNDI is a useful alternative to LOS, but future work may modify LOSNDI to become a more sensitive outcome variable, especially in patients not having complications. The SF-36 QoL questionnaire did not reveal any convincing significant effect of nutritional intervention. The response rate was much lower than in the study of Beattie et al.15 in which all patients responded to both questionnaires. In their study of patients undergoing gastrointestinal or vascular surgery, both the mental score and the physical score improved significantly. Response rate is known to be influenced by the severity of disease, and quality of life is more likely to improve when recovering from an acute disease compared to a chronic conditioning.18 It seems that the SF-36 quality of life questionnaire is not suitable as an outcome variable in a mixed population of patients. This study is, to the best of our knowledge, the first attempt to validate a nutritional screening system in a randomized controlled trial. Future work will clarify whether other effects can be discovered in the large part of the at-risk patients who did not have complications. However, with the present results there is a rationale for initiating nutritional support in all patients identified as being nutritionally at risk since those who develop complications will have a significant clinical benefit from this.

ARTICLE IN PRESS 550

Acknowledgements The study was financed by the Danish Ministry of Health and the hospitals participating in conjunction with the County of Nordjylland, the County of Storstrom and the Copenhagen Hospital Corporation Organisation (H:S).

References 1. McWhirter JP, Pennington CR. Incidence and recognition of malnutrition in hospital. BMJ 1994; 308(6934):945–8. 2. Kelly IE, Tessier S, Cahill A, et al. Still hungry in hospital: identifying malnutrition in acute hospital admissions. QJM 2000;93(2):93–8. 3. Edington J, Boorman J, Durrant ER, et al. Prevalence of malnutrition on admission to four hospitals in England. The malnutrition prevalence group. Clin Nutr 2000; 19(3): 191–5. 4. Corish CA, Flood P, Mulligan S, Kennedy NP. Apparent low frequency of undernutrition in Dublin hospital in-patients: should we review the anthropometric thresholds for clinical practice? Br J Nutr 2000; 84(3):325–35. 5. Windsor JA, Hill GL. Risk factors for postoperative pneumonia. The importance of protein depletion. Ann Surg 1988; 208(2):209–14. 6. Jeejeebhoy KN. Muscle function, nutrition. Gut 1986; 27(Suppl 1):25–39. 7. Giner M, Laviano A, Meguid MM, Gleason JR . In 1995 a correlation between malnutrition and poor outcome in critically ill patients still exists. Nutrition 1996;12(1): 23–9. 8. Stratton RJ, Green CJ, Elia M. Disease-related malnutrition: an evidence-based approach to treatment. Wallingford UK: CAB International; 2003. 9. Kondrup J, Rasmussen HH, Hamberg O, Stanga Z. Nutritional Risk Screening (NRS 2002): a new method based on an analysis of controlled clinical trials. Clin Nutr 2003; 22(3):321–36.

N. Johansen et al.

10. ESPEN Education and Clinical Practice Committee. ESPEN Guidelines for Nutrition Screening 2002. Clin Nutr 2003; 22(4):415–21. 11. Kondrup J, Johansen N, Plum LM, et al. Incidence of nutritional risk and causes of inadequate nutritional care in hospitals. Clin Nutr 2002; 21(6):461–8. 12. Pedersen AN, Ovesen L. Recommendations for food in public institutions in Denmark. Copenhagen: Danish Ministry of Food and Agriculture; 1999. 13. Elmore MF, Gallagher SC, Jones JG, Koons KK, Schmalhausen AW, Strange PS. Esophagogastric decompression and enteral feeding following cholecystectomy: a controlled, randomized prospective trial. J Parenter Enteral Nutr 1989; 13(4):377–81. 14. Cabre! E, Gonzalez-Huix F, Abad-Lacruz A, et al. Effect of total enteral nutrition on the short-term outcome of severely malnourished cirrhotics. A randomized controlled trial. Gastroenterology 1990; 98(3):715–20. 15. Beattie AH, Prach AT, Baxter JP, Pennington CR. A randomised controlled trial evaluating the use of enteral nutritional supplements postoperatively in malnourished surgical patients. Gut 2000; 46(6):813–8. 16. Buzby GP, Knox LS, Crosby LO, et al. Study protocol: a randomized clinical trial of total parenteral nutrition in malnourished surgical patients. Am J Clin Nutr 1988; 47(Suppl 2):366–81. 17. Garner JS, Jarvis WR, Emori TG, Horan TC, Hughes JM. CDC definitions for nosocomial infections. Am J Infect Control 1988; 16:128–40. 18. Bjrner JB, Damsgaard MT, Watt T, et al. Danish manual in SF-36, a questionnaire on health status. Copenhagen: LiF, 1997. 19. Ware JE, Jr., Sherbourne CD. The MOS 36-item short-form health survey (SF-36). I. Conceptual framework and item selection. Med Care 1992; 30(6):473–83. 20. Altman DG. Practical statistics for medical research. London: Chapman & Hall; 2002. 21. Kondrup J. Can food intake in hospitals be improved? Clin Nutr 2001; 20(Suppl 1):153–60. 22. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1(8476):307–10.