TIAMAT- towards an interdisciplinary automated malnutrition screening tool

Area of Science:

• Clinical Nutrition
• Medical Informatics
• Biochemistry

Background:

• Hospital malnutrition is a significant issue, with current screening methods like Nutrition Risk Screening (NRS-2002) facing challenges in routine implementation.
• Automated, staff-independent approaches are needed to improve malnutrition detection and management.
• Standard lab tests like serum albumin are insufficient, but combinations may predict malnutrition risk.

Purpose of the Study:

• To develop and validate an automated malnutrition screening tool (TIAMAT) using routine laboratory data.
• To compare the performance of the new tool against existing methods (NRS-2002, MUST).
• To assess the potential for integrating lab-based screening into routine hospital data viewing.

Main Methods:

• A cohort of 300 internal medicine patients was studied.
• Data from NRS-2002, MUST, and Subjective Global Assessment (SGA) were collected.
• A multivariate logistic model was developed using a training set (n=200) to predict SGA from laboratory results, forming the TIAMAT score.
• The model was validated on a separate set (n=100), with missing data imputed.

Main Results:

• The TIAMAT score in the training set showed sensitivity of 71% and specificity of 81% for predicting SGA.
• In the validation cohort, TIAMAT achieved 60% sensitivity and 75% specificity.
• Performance metrics for TIAMAT were comparable to MUST and NRS-2002, with AUCs of 0.77 (TIAMAT) vs 0.80 (MUST) and 0.92 (NRS-2002).

Conclusions:

• Combinations of standard laboratory tests can form an effective alternative to current malnutrition screening methods.
• Automated display of these results could facilitate universal screening across hospital populations.
• This approach has the potential to significantly improve the efficacy of hospital-wide malnutrition screening.