A simplified multidimensional scale approach is effective in predicting mortality in hospitalized older adults and highlights the role of nutrition

Abstract

Background & aims

Malnutrition and cognitive impairment are among the major contributors to frailty, that significantly increases the risk of mortality of older hospitalized patients. Multidimensional frailty assessment tools, such as the multidimensional prognostic index-MPI, a tool based on a standard comprehensive geriatric assessment (CGA), have proven valuable for predicting adverse outcomes, including mortality of older adults following acute illness but its application in everyday clinical practice is limited. We hypothesized that removing parameters not closely associated with mortality and sorting the patient population according to the presence or not of cognitive impairment with possible integration of common laboratory markers, could provide a simplified approach that could improve practicability in all settings with at least comparable 1-year mortality predictive value.

Methods

A retrospective cohort study was conducted in patients consecutively admitted to the Geriatric Clinic of the Maggiore University Hospital in Trieste, Italy from January 1st 2018 to December 31st 2019. Their demographics, functional, clinical, laboratory parameters and 1-year mortality were recorded. In a development cohort of 1032 consecutive patients, best predictors of mortality were selected via systematic analysis and included in simplified prognostic models and algorithms and subsequently compared for prediction of 1-year mortality. The predictive relevance of the best algorithms was then validated, in comparison to MPI, in a separate cohort of 575 consecutive patients.

Results

While all demographic and tested laboratory parameters as well as MPI domains correlated with 1-year mortality, exclusion from MPI calculation of Short Portable Mental Status Questionnaire (SPSMQ), Exton Smith scale (ESS) and Mini Nutritional Assessment (MNA) significantly reduced MPI mortality predictivity, suggesting that not all MPI domains have the same weight. Further analysis showed that in the whole study cohort and in subgroups according to cognitive function, selected models including up to 3 parameters were superior to MPI in predicting 1-year mortality. In particular, models including MNA and albumin, or Exton Smith scale proved to better predict mortality in patients without or with severe cognitive impairment, respectively. A derived diagnostic algorithm applying different models according to cognitive status showed improved predictive value compared to MPI while requiring shorter estimated assessment time. Internal validation confirmed these results [HR: 4.37 (3.02–6.31) vs 3.16 (2.18–4.61), p < 0.0001].

Conclusions

In older acutely ill patients, a simplified multidimensional algorithm approach based on the assessment of cognitive function followed by nutritional status with the addition of plasma albumin or of functional status in patients without or with severe cognitive impairment respectively, may significantly improve 1-year mortality prediction while reducing assessment time. Moreover, these results highlight the prognostic value of MNA in association with albumin for 1-year mortality risk screening in the hospital setting and, for the first time, demonstrate its differential performance according to the presence or not of cognitive impairment.