Multicentre prospective study on the diagnostic and prognostic validity of malnutrition assessment tools in surgery

2025-03-04

Our latest research study was published open access in the prestigious British Journal of Surgery (Incorporating the European Journal of Surgery), on 3 March 2025.

Multicentre prospective study on the diagnostic and prognostic validity of malnutrition assessment tools in surgery

Georgia Petra, Evangelos I Kritsotakis, Nikolaos Gouvas, Dimitrios Schizas, Konstantinos Toutouzas, Michael Karanikas, George Pappas-Gogos, Georgios Stylianidis, George Zacharioudakis, Aggelos Laliotis, Grigorios Christodoulidis, Ioannis Kehagias, Konstantinos Lasithiotakis,

on behalf of the MATS Study Group

Access the full paper via:

Abstract

Background

Malnutrition is a risk factor for postoperative morbidity but the optimal tool for the assessment of malnutrition is unclear.

Methods

This is a prospective multicentre cohort study. Consecutive patients undergoing elective or emergency major abdominal surgery for benign or malignant disease in 12 Greek hospitals between January 2022 and December 2023 were included. Patients unable to provide nutrition history and/or informed consent were excluded. Subjective global assessment (SGA) was used as a reference standard for malnutrition diagnosis. GLIM (global leadership initiative on malnutrition), MNA-SF (mini nutrition assessment short form), MST (malnutrition screening tool), MUST (malnutrition universal screening tool), NRI (nutritional risk index), NRS-2002 (nutrition risk scale 2002), PONS (perioperative nutrition screen) and SNAQ (short nutrition assessment questionnaire) tools were applied for malnutrition risk assessments. Indicators of diagnostic accuracy (sensitivity, specificity, diagnostic odds ratio, areas under the receiver operating characteristic curve—AUC), construct validity (convergent associations with relevant variables) and prognostic validity (logistic regression) were appraised.

Results

1649 patients were included (58% colorectal, 21% upper gastrointestinal, 14% hepatobiliary operations). SGA defined 562 (34.1%) patients as malnourished with excellent construct and prognostic validity. Malnutrition risk assessments varied from 24.0% using NRS-2002 to 58.6% with the MNA-SF. On their ordinal scales, MNA-SF (AUC = 0.83, 95% c.i. 0.81 to 0.85) and MUST (AUC = 0.79, 95% c.i. 0.77 to 0.82) had the best discriminatory abilities with minimal between-centre heterogeneity. As binary classifiers, MNA-SF (OR = 30.2; 95% c.i. 20.2 to 45.1) and MUST (OR = 16.1; 95% c.i. 12.4 to 21.1) had the highest diagnostic ORs but only MUST had sensitivity and specificity close to 80%. MUST performed well in construct and prognostic validity appraisals.

Conclusion

This study supports the use of the MUST as it is the most valid nutritional screening tool in patients after major abdominal surgery.

Statistical methods

A precision-based calculation determined that 1437 patients were required to ensure, with probability 0.95, that anticipated 80% sensitivity and 80% specificity to detect malnutrition would be estimated with precision ±4% in a 95% confidence interval. The calculation assumed an underlying true malnutrition prevalence of 30% and was based on Student's t distribution (rather than the Normal) to incorporate uncertainty about anticipated estimates16. Because loss of precision may occur due to intraclass correlation (that is, data from the same hospital are more similar than those from different hospitals), the sample size was inflated by 15%, assuming a design effect of 1.15 (intraclass correlation 0.10) and moderately heterogeneous distributions among hospitals. The aim was to recruit 1650 patients in the study.

A comprehensive comparative assessment of the nutritional screening tools was pursued by examining measures of concurrent, construct and prognostic validities. Concurrent validity was examined using prevalence-independent measures of diagnostic accuracy (sensitivity, specificity and diagnostic odds ratio) for binary classification of malnutrition, employing SGA as the reference standard. In keeping with common clinical practice, a dichotomous version of each screening tool was examined, with two levels: 'at risk' (combining categories for moderate and high risk) and 'low risk' of malnutrition. Similarly, SGA was classified into 'malnourished' and 'well nourished'. Confidence intervals for sensitivity and specificity were computed using Wilson's score method, whereas confidence intervals for diagnostic odds ratios based on a log-normal distribution. A priori, concurrent validity was defined as adequate when both sensitivity and specificity were 80% or higher.

In addition, the overall and centre-specific areas under the receiver operating characteristic curve (AUC) were estimated using the original ordinal scale of each screening tool. To examine interhospital heterogeneity, a random effects model was applied, assuming a Normal between-hospital distribution for the logit transformation of the AUC. A Wald-type 95% c.i. was calculated to indicate highly probable values of pooled AUC for each screening tool. A 95% prediction interval was calculated to indicate the heterogeneity expected in the AUC when a screening tool is applied to a new population. AUCs above 0.80 were considered acceptable discriminative ability.

Examination of construct validity was based on the convergent associations of each screening tool with variables it should be associated with, if the tool sufficiently identifies malnutrition. Cross-sectional associations between malnutrition risk ratings from each tool and known medical, physical and functional correlates of malnutrition were looked for. The associations were quantified using odds ratios estimated by fixed-effects logistic regression and were converted into Cohen's d standardized effect sizes to aid with interpreting the results.

Prognostic validity for each malnutrition screening tool was evaluated by examining the strength of its association with hospital mortality rate and the occurrence of serious 30-day postoperative complications. ORs for each outcome were estimated by mixed-effects logistic regression adjusting for known preoperative prognostic factors, including sex, age, ASA class, ambulatory status, emergency admission, operative severity, sepsis 48 h before surgery, ascites 30 days before surgery and history of diabetes mellitus. A random intercept term was included to account for between-hospital effects. Prognostic validity was considered substantial when OR > 2.

MUST and NRI could not be calculated for 35 (2%) patients, for whom weight loss could not be quantified. There were no missing data for other variables and no imputation method was applied. The analyses involving MUST and NRI were performed on a complete case basis (listwise deletion). Two-sided confidence intervals and P were reported, considering statistical significance when P < 0.05. Stata version 18 (Stata Corporation, College Station, Texas, USA) was used for all analyses.

Fig. 3. Prognostic validity of preoperative malnutrition risk assessments for predicting hospital mortality rate (left panel) and 30-day postoperative major complications (right panel)

Fig. 2. Forest plot of the pooled area under the receiver operating characteristic curve (AUC) with 95% c.i. and prediction intervals from random-effects meta-analysis of hospital-specific data

Fig. 1. Forest plot of the variation of malnutrition risk assessments by nutritional screening tools

Table: Indicators of concurrent criterion validity. Data are point estimates with respective 95% c.i. reported in parentheses. The binary form of each screening tool was used to calculate the diagnostic accuracy measure using the Subjective Global Assessment (SGA) as a reference.