Nutritional support in hospitalised patients with diabetes and risk for malnutrition: a secondary analysis of an investigator-initiated, Swiss, randomised controlled multicentre trial

Study design and setting

This is a secondary analysis of the prospective, investigator-initiated, randomised controlled Swiss multicentre trial (EFFORT)19 that enrolled medical inpatients at nutritional risk. The trial studied the effect of individualised nutritional support compared with usual hospital nutrition on clinical outcomes including mortality in eight secondary and tertiary care hospitals throughout Switzerland. Its protocol, the main results and several secondary analyses have previously been published.19 22–28

Study population

The initial assessment concerning nutritional risk was done using the Nutritional Risk Screening 2002 (NRS).29 The NRS is a validated screening tool to assess the risk of malnutrition in the in-hospital setting. The tool is comprised of several subcategories to which a predefined number of points (0–3) is attributed. These categories are the severity of underlying disease (0–3 points), nutritional status, divided into body mass index (BMI), weight loss and food intake (0–3 points), plus one point for age ≥70 years (maximum seven points). The higher the patients score, the higher the risk for malnutrition. Patients with an NRS total score of 3 or greater that were expected to stay at least 4 days in the hospital were included. Exclusion criteria were an initial hospitalisation in intensive care units or surgical wards; unable to ingest oral nutrition; already receiving nutritional support at admission; with a terminal condition; admitted to hospital because of anorexia nervosa, acute pancreatitis, acute liver failure, cystic fibrosis or stem cell transplantation; after gastric bypass surgery; with contraindications for nutritional support; and previously included in the trial. For this secondary analysis, we included patients with a previous diagnosis of diabetes and those that met the diagnosis of diabetes based on Haemoglobin A1c (HbA1c) value ≥6.5%.30

Study intervention

After inclusion in the trial, patients were randomised 1:1 using an interactive web system. Patients in the intervention group received individualised nutritional support within 48 hours after admission. The nutritional support was based on a previous consensus protocol and in agreement with guidelines for polymorbid patients.22 31 A registered, trained dietitian developed a personalised nutritional plan. The daily protein target was set at 1.2–1.5 g/kg of total bodyweight (0.8 g/kg of total bodyweight for patients with renal failure (estimated glomerular filtration rate (eGFR) <30 mL/min)) and the Harris-Benedict equation, adjusted for activity and severity of disease, was used to establish energy requirements.32 An oral nutrition plan was developed, which could potentially be extended to enteral tube feeding and parenteral nutrition if the protein and caloric goals were not reached within 5 days. There was no specific protocols regarding nutritional support for patients with diabetes. Nutritional intake was re-assessed every 24–48 hours. Patients in the control group received standard hospital nutrition without nutritional counselling or additional support.

Endpoints

The primary endpoint for this analysis was all-cause mortality within 30 days, which corresponds to the principal secondary endpoint of the initial EFFORT analysis.22 Secondary endpoints included adverse clinical outcomes such as admission to an intensive care unit, non-elective readmission after discharge, length of stay or decline in functional status measured by a decrease of at least 10% as measured by the Barthel index (primary endpoint of the initial trial). Furthermore, we included major complications such as respiratory insufficiency, nosocomial infections, acute kidney injury, gastrointestinal failure and major cardiac events. Safety endpoints were severe hyperglycaemia, occurrence of refeeding syndrome, complications related to enteral or parenteral nutrition, liver or gallbladder dysfunction and adverse gastrointestinal effects. A detailed description of outcomes has been published previously.22

Structured follow-up interviews to assess outcomes were conducted by telephone by a blinded study nurse on day 30 and day 180 and annually thereafter. Mortality during follow-up was verified by family members or the patient’s primary care physician.

Statistical analysis

Continuous variables are expressed as mean±SD; binary and categorical variables as number or count and percentage. Statistical significance was tested at 95% CIs, corresponding to a p value of 0.05.

For the main analysis, we investigated the effect of nutritional support on mortality and other secondary endpoints in patients with diabetes and nutritional risk. We used Cox regressions for all mortality analyses and report HRs, as well as Kaplan-Meier estimates to graphically display survival rates. For other binary endpoints, we used logistic regression models and report ORs. Similar to the initial trial analysis,19 22 models were adjusted for predefined covariates including study centre, nutritional risk (based on the NRS total score) and baseline Barthel index. For all subgroup analyses, we calculated an interaction analysis with reporting of p for interaction to understand whether the effect within patients with diabetes would be different from the overall trial population.

In the second step, we investigated the association between diabetes and clinical outcomes in the overall trial population (n=2028). We used the same statistical approach as defined above. Further, we analysed the association of nutritional risk, based on the NRS 2002 score, and long-term mortality within 180 days within the diabetic population.

To understand the effects of nutritional support on glucose control, means of blood glucose values of all patients and each point of measurement were calculated. We had information on available blood glucose data taken as part of standard clinical care of 405 out of 445 patients with diabetes taken on 6 consecutive days. Measurements were done pre-prandially in the morning, at noon and in the evening. There was considerable interpatient variance concerning the number of measurements per patient. Cumulative 6-day mean values were calculated for each patient and each time of measurement. Due to deviation from the normal distribution of blood glucose data, Wilcoxon-Mann-Whitney tests were used to evaluate the differences in glucose serum levels between the intervention and the control groups. We compared the frequency and the rate of hyperglycaemic events between the intervention and the control groups of patients with diabetes. Hyperglycaemia was defined as blood glucose >7 mmol/L.

All statistical analyses were performed with STATA V.17 (Stata Corp, College Station, Texas, USA).