Predictors of incident malnutrition—a nutritionDay analysis in 11,923 nursing home residents

Nursing home (NH) residents number approximately 1.6 million in the US (2015) and 4.0 million in Europe (2017). Malnutrition is common in NH residents, reported in up to 54% depending on definitions and cut-offs, and is linked to poor health and functional outcomes and higher healthcare costs. Preventing malnutrition is therefore crucial. However, little is known about the development of malnutrition over time in NH residents. Cross-sectional studies cannot establish directionality, and existing longitudinal studies have been few with small samples and often included residents already malnourished at baseline, complicating incidence estimates. This study aims to identify predictors of incident malnutrition over 6 months among older NH residents, focusing on modifiable and non-modifiable risk factors to inform prevention.

Prior cross-sectional research has associated depression, poor oral intake, swallowing disorders, chewing problems, eating dependency, immobility, dependence in activities of daily living (ADL), and female sex with weight loss (WL) or low BMI in NH settings. Four longitudinal studies reported cognitive decline, lower functional status/higher ADL dependency, constipation, hospitalization, eating dependency, and low appetite as risk factors for malnutrition, but were limited by small sample sizes (n=108–441) and often did not exclude those malnourished at baseline. Repeated-measures data suggest that 23% of NH residents experienced a decline in nutritional status over one year. These gaps underscore the need for large-scale longitudinal analyses excluding malnourished individuals at baseline and including additional potential risk factors such as low food intake.

Reporting: The study adheres to STROBE guidelines for observational studies. Study design: Secondary analysis within the Joint Action MaNuEL (JPI-HDHL) using worldwide nutritionDay (nD) data from NHs collected annually with a 6-month follow-up (FU). Ethical approvals were obtained; participants or legal representatives consented; data were anonymized. Participants: NH residents aged ≥65 years participating between 2007–2018. Exclusions: no FU body weight; residents from Japan; missing baseline BMI or pre-nD weight loss (WL) data; malnutrition at baseline (BMI <20 kg/m² and/or WL >5 kg in past year); missing age/sex or variables with <0.1% missing without a missing category. Final analytic sample: 11,923 residents from North America and Europe. Data acquisition: Local NH staff completed standardized questionnaires (>30 languages). Weight measured on ward scales at nD and FU or taken from records. Height measured with stadiometer if possible, otherwise estimated from knee height or records. BMI calculated as kg/m² using height at nD for both time points. Outcome: Incident malnutrition defined as BMI <20 kg/m² and/or WL ≥10% between nD and FU (difference in body weight between time points). Potential predictors (17 variables): - General characteristics: age (65–74, 75–84, 85–94, 95–107), sex, BMI quartiles (20.0–22.9, 23.0–25.6, 25.7–29.0, 29.1–64.8 kg/m²). - Function: mobility (mobile, partially mobile, immobile), cognitive impairment (none, slight–moderate, severe; per staff assessment or tests), dysphagia (yes/no), chewing problems (yes/no). - Nutrition: intake at lunch (three-quarters to all; half; a quarter; nothing; nothing due to artificial nutrition). - Diseases: cancer (yes/no), neurologic (e.g., dementia, stroke) (yes/no), musculoskeletal (e.g., rheumatoid arthritis, osteoporosis) (yes/no), cardiovascular/pneumological (yes/no), other diseases (yes/no). - Medication: antibiotics (yes/no), opiates (yes/no), psychoactive substances (yes/no), number of drugs (<5/≥5). Statistical analysis: Descriptive statistics for predictors. For variables with >0.1% missing, a missing category was added; cases with <0.1% missing in specific variables (tube/parenteral nutrition, cancer) and missing age/sex were excluded. Associations with incident malnutrition were assessed with generalized estimating equations (GEE) clustering by NH ward. Each predictor was first tested in univariate GEE; those with p<0.1 were checked for multicollinearity (VIF<5) and entered into multivariate GEE. Final model significance threshold p<0.005; potential interactions by sex tested. Results reported as odds ratios (ORs) with 99.5% confidence intervals (99.5% CI). Complete case sensitivity analysis performed. Model discrimination summarized by AUC. Analyses in R 3.6.1.

Sample flow: Of 39,840 residents assessed at baseline (nD), after exclusions (no FU weight, age <65, Japan, missing baseline WL/BMI, baseline malnutrition, minimal missing data), 11,923 non-malnourished residents were analyzed. Incidence: At 6 months, 953 (8.0%) had WL ≥10%, 586 (4.9%) had BMI <20 kg/m², and 290 (2.4%) met both criteria; overall, 1,249 (10.5%) developed incident malnutrition. Predictors (multivariate GEE, 13 variables included; model AUC 0.68): - Intake at lunch (vs three-quarters to all): nothing OR 2.79 (99.5% CI 1.56–4.98); a quarter OR 2.15 (1.56–2.97); half OR 1.72 (1.40–2.11). - BMI quartiles (vs highest 29.1–64.8): lowest 20.0–22.9 OR 1.86 (1.44–2.40). - Age group (vs 65–74): 85–94 years OR 1.46 (1.05–2.03). - Cognitive impairment (vs none): severe OR 1.38 (1.04–1.84). - Mobility (vs mobile): immobile OR 1.28 (1.00–1.62). Univariate associations: All predictors except tube feeding, parenteral nutrition, cancer, cardiovascular/pneumological diseases, other diseases, and antibiotics had p<0.1. No multicollinearity detected. Sex interaction significant only for BMI 20.0–22.9, with higher odds for males than females in this subgroup. Sensitivity analysis (complete cases n=11,170): Similar effect sizes; immobility lost statistical significance.

This large international cohort of non-malnourished NH residents shows that 10.5% develop malnutrition over 6 months, addressing the need to understand incident cases rather than prevalence in mixed-status samples. The strongest predictor was reduced intake at lunch, with a graded increase in risk as intake decreased, supporting the role of current nutritional intake as a proximate determinant of subsequent WL/low BMI. Low baseline BMI close to the diagnostic threshold plausibly increased incident malnutrition risk. Severe cognitive impairment and immobility also predicted incident malnutrition, consistent with pathways involving mealtime difficulties, reduced self-feeding ability, posture-related intake problems, and behavioral issues that limit intake. Residents aged 85–94 had higher risk compared to 65–74, suggesting increased vulnerability in very old age groups. Comparisons with prior studies are complicated by differing definitions and inclusion of malnourished participants at baseline in earlier work; nevertheless, the current findings align with associations reported for cognitive impairment, poor functional status, and low appetite/intake. The reliance on lunch intake as a proxy for overall intake is supported by prior hospital data and the centrality of lunch as the main warm meal in many countries. The observed associations underscore modifiable targets (nutritional intake support, functional and cognitive interventions) and highlight groups for prioritized monitoring and preventive care. The overall model discrimination (AUC 0.68) indicates moderate ability to identify high-risk individuals using routinely available data.

In this multi-country nutritionDay cohort, 6-month incident malnutrition among initially non-malnourished nursing home residents was 10.5%. Key predictors were low intake at lunch, low BMI (20.0–22.9 kg/m²), severe cognitive impairment, immobility, and age 85–94 years. These findings emphasize the need to monitor meal intake and target nutritional care to high-risk residents, alongside interventions addressing functional and cognitive limitations. Future research should include longer follow-up with repeated nutritional assessments, detailed characterization of reasons for weight loss and low intake, better documentation of drop-out reasons, and evaluation of whether incorporating these predictors into screening tools improves care processes and outcomes.

- Data collection by local NH staff using routine methods rather than trained research personnel or validated scales (e.g., Katz/Barthel for ADL) may introduce measurement error. - The nD dataset constrained available variables; potentially relevant predictors (e.g., ADL scores, multimorbidity indices, hospitalization) were unavailable. - Food intake was estimated via a plate diagram for lunch only, not through comprehensive dietary assessment or across all meals. - High proportion of residents without 6-month follow-up weight and exclusions for missing baseline data may introduce selection bias and limit external validity; incidence could be underestimated if those lost to follow-up were at higher risk. - Inclusion limited to NHs/wards with follow-up may bias toward facilities more engaged in nutrition care, potentially underestimating incidence. - As an observational study, residual and unmeasured confounding cannot be excluded. - In complete case sensitivity analysis, the immobility association lost statistical significance, possibly due to reduced power.