Evaluation of Nutritional Support and In-Hospital Mortality in Patients With Malnutrition

This cohort study examines the association of nutritional support with mortality rates among hospitalized patients with malnutrition in Switzerland.


Introduction
Malnutrition is defined as a state of insufficient intake or uptake of nutrients, leading to an altered body composition. 1 As a risk factor associated with adverse outcomes, it is diagnosed in a considerable proportion of patients in hospitals. 2,3 Among patients in hospitals in Switzerland, a 2010 study 4 found a malnutrition prevalence rate of 18.2% and a 2015 study 2 found a rate of 27.8%, using the nutritional risk screening (NRS 2002) score. 5 Patients who are older and considered frail have high rates of illness-related impaired protein and energy homeostasis, hormonal changes, and reduction of appetite associated with disease-related malnutrition. 1,6 Importantly, the association between illness and nutritional status seems to be bidirectional, with malnutrition also associated with morbidity, mortality, functional decline, prolonged hospital stays, and higher health care costs. 2,7 In light of the high number of patients with malnutrition in the hospital, awareness of the detrimental associations of malnutrition among patients with multiple conditions has increased. 8,9 While a 2016 study 10 did not find an association between nutritional interventions and mortality or other clinical outcomes, randomized clinical trials from 2019 11 and 2016 12 found that nutritional support reduced risks for mortality and complications and improved functional outcomes and quality of care. Therefore, current practice guidelines issued by the European Society for Clinical Nutrition and Metabolism 3 and the American Society for Parenteral and Enteral Nutrition 13 suggest screening for malnutrition, nutritional assessment, and nutritional support for inpatients with malnutrition.
An increasing number of hospitals are implementing strategies for more standardized screening and management of patients with malnutrition. As such, we aimed to use a large Swiss nationwide claims database to assess whether the administration of nutritional support in clinical routine was associated with improved in-hospital survival compared with receiving no nutritional support.

Study Design
The institutional review board of Northwestern Switzerland approved this cohort study and waived the requirement of participant informed consent, as the data were deidentified. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
We conducted a nationwide population-based cohort study using an administrative claims database (ie, Medizinstatistik) provided by the Federal Statistical Office in Switzerland. The database contains longitudinal, individual-level data on in-hospital health care use, inpatient diagnoses, diagnostic tests, and procedures. It includes all Swiss inpatient discharge records from acute care, general, and specialty hospitals, excluding hospital units of postacute care institutions, regardless of payer. The database thus provides a near-complete sample of hospitalizations in Switzerland. Each hospitalization is identified uniquely in the database, so rehospitalizations could be tracked. As data are assessed for every hospital stay, a single patient may have more than 1 index admission in our cohort. 14 covariates were selected a priori on the basis of clinical experience, expertise, and published literature. 17

Data Sources and Study Population
We included hospitalizations of adult (age Ն18 years) patients with malnutrition and noncritical illness in medical wards. Hospitalizations with any intensive care unit (ICU) admission were excluded to increase specificity. Medical diagnoses were coded using the ICD-10, German Modification (ICD-10-GM) codes. 18 Main diagnoses consisted of the main reason for hospitalization and in-hospital care; secondary diagnoses were a measure for the level of comorbidity. Hospitalizations of patients with a diagnosis of malnutrition were identified using ICD-10-GM codes E43, for unspecified severe protein-energy malnutrition; E44, for protein-energy malnutrition of moderate or mild degree; and E46, for unspecified protein-energy malnutrition (eTable 1 in the Supplement). The following codes were not considered, as they are not common reasons for malnutrition in Swiss hospitals: E40, for kwashiorkor; E41, for nutritional marasmus; E42, for marasmic kwashiorkor; and E45, for retarded development following protein-energy malnutrition.
The study period was from April 2013 to December 2018. In 2012, the Swiss Diagnosis-Related Groups (SwissDRG) reimbursement system was implemented, and since then, coding for malnutrition and nutritional support has gained financial relevance, as reimbursement for hospitalizations with codes for malnutrition is now higher than those without such codes. While encoding rules for malnutrition were relevantly adapted by SwissDRG in March 2013 by basing the ICD-10 coding mainly on the criteria of the NRS 2002, 19 only minor changes in coding have been made since then. 20 The NRS-2002 score assesses the risk for malnutrition and differentiates between patients at nutritional risk (ie, those with Ն3 points) and patients with malnutrition (ie, those with Ն4 points). In this study, we used the term malnutrition for all patients with an NRS score of 3 or greater, in accordance with a 2019 clinical trial. 11 From a total of 1 892 131 adults who were hospitalized during our study period, we identified 114 264 patients (6%) who were eligible based on inclusion and exclusion criteria (Figure 1).

Exposure
To assess the associations of nutritional support with clinical outcomes of interest, all hospitalizations of patients with malnutrition were screened for the presence of nutritional support based on CHOP codes. The following CHOP codes were counted as nutritional support: 89.0A.32 and 89.0A.4X, for dietary advice and nutritional therapy; 96.6, for enteral infusion of concentrated nutrients; and 99.15, for parenteral infusion of concentrated nutrient solutions.

Outcomes
The primary outcome was all-cause in-hospital mortality. Secondary outcomes were all-cause 30-day readmission rate and discharge to a postacute care facility. For disposition status, the event of interest was discharge to a short-term or long-term postacute care facility or discharge home.

Statistical Analysis
To control for confounding by indication, we fitted different Poisson regression models for in-hospital all-cause mortality and 30-day readmission, and results were reported as incidence rate ratios (IRRs) with their respective 95% CIs. Similarly, we reported odds ratios (ORs) for discharge to a postacute care facility using logistic regression models. Comparing hospitalizations of patients receiving nutritional support with hospitalizations of patients not receiving nutritional support, we performed unadjusted and multivariable regression analyses, adjusting for sociodemographic factors (ie, age, sex, nationality, insurance status, month and year of admission, mode of admission, location before admission, hospital size, and hospital site). The fully adjusted model additionally included main diagnoses, comorbidities, severity of malnutrition, total number of hospitalizations, use of palliative treatment, Charlson Comorbidity Index, hospital frailty risk score, and hospital LOS. We performed a propensity score-matched analysis. Eligible hospitalizations of patients receiving nutritional support were 1:1 propensity score matched to a comparative general medical cohort of patients in the hospital with malnutrition not receiving nutritional support (ie, the matched control population). The probability of receiving nutritional support vs not receiving nutritional support was calculated via a multivariable logistic regression model that contained all baseline covariates and month of admission. No data were missing in our study. The estimated propensity score was used to match hospitalizations of patients with malnutrition receiving nutritional support with patients in a nearest-neighbor control group using a caliper size of 0.0005 on the propensity scale. Covariate balance before and after propensity score matching was assessed using standardized differences. A standardized difference less than 10% was interpreted as indicating adequate balance between groups. 21 After propensity score matching, estimates of the effect sizes and corresponding 95% CIs were determined using Poisson or logistic regressions as appropriate; the models were additionally adjusted for hospital site. Kaplan-Meier curve was used to illustrate differences in time to in-hospital mortality, and Cox proportional hazards regression was used for the calculation of hazard ratios (HRs).

JAMA Network Open | Nutrition, Obesity, and Exercise
We performed sensitivity analyses of patients who were treated with only dietary advice or oral nutritional support (ie, excluding patients receiving enteral or parenteral nutrition). For exposed and unexposed groups, we calculated a separate propensity score (using a caliper size of 0.0005) and performed a separate 1:1 matching.  Because in-hospital mortality functions as competing risk associated with the 2 secondary outcomes, we analyzed competing event data using the cumulative incidence function and performed the nonparametric Gray test to compare these functions. 22 To explore the consistency of the findings among subgroups, we included interaction terms in the regression models to test for modification associated with baseline factors. Specifically, we tested for subgroups by patient age, sex, main diagnosis during hospitalization, comorbidities, degree of malnutrition, hospital frailty risk score, and hospital LOS.
All P values were 2-sided and were not adjusted for multiple testing, and P < .05 was considered statistically significant. All statistical analyses were performed using Stata statistical software version 15.1 (StataCorp). Data were analyzed from February 2020 to May 2020.

Results
Among 69 934 patients with malnutrition who were hospitalized and propensity score matched, the mean (SD) age was 73.8 (14.5)

Primary Outcome
Among 77 460 patients who were unmatched and receiving nutritional support, 6633 patients (8.6%) died during hospitalization, compared with 3239 patients (8.8%) who died among 36 804 patients who were unmatched and not receiving nutritional support. This outcome corresponded to an unadjusted IRR of 0.77 (95% CI, 0.74-0.80; P < .001) for in-hospital mortality associated with receiving nutritional support vs not receiving nutritional support. We also found that nutritional support was significantly associated with reduced mortality rate in the fully adjusted model, with an IRR of 0.76 (95% CI, 0.73-0.80; P <.001).
Among 69 934 patients who were propensity score matched, 5597 patients died, for an overall all-cause in-hospital mortality rate of 8.0%. Among matched patients, 2525 of 34 967 patients receiving nutritional support died, for a mortality rate of 7.2%, compared with 3072 of 34 967 patients who died in the matched control group, for a mortality rate of 8.8% (IRR, 0.79 [95% CI, 0.75-0.84]; P < .001) ( Table 2). Kaplan-Meier survival curve showed lower cumulative incidence of in-hospital deaths over 30 days for the group receiving nutritional support, with an adjusted HR of 0.79 (95% CI, 0.75-0.84; P < .001) (Figure 2).

Secondary Outcomes
In the unadjusted analysis, we found significantly lower 30-day readmission rates among patients

Sensitivity and Subgroup Analyses
Among patients receiving nutritional support only (ie, when excluding patients receiving enteral or parenteral nutrition), after matching, the IRR for overall mortality was 0.75 (95% CI, 0.71-.080; P < .001) compared with patients not receiving nutritional support (eTable 2 in the Supplement).
Using the propensity score-matched population, we performed several subgroup analyses stratifying for main baseline covariates (Figure 3). In general, our findings for in-hospital mortality remained robust, except for the following conditions: among patients who were hospitalized with a lower hospital frailty risk score (ie, <5 points), receiving nutritional support was associated with a greater reduction in mortality risk compared with patients with higher frailty risk scores. Among patients with a shorter hospital LOS (ie, <5 days), prescribed nutritional intervention was associated with a greater decrease in mortality risk compared with those who were hospitalized more than 5 days.

Discussion
This large population-based cohort study of more than 110 000 patients with malnutrition in hospitals had 2 key findings. First, we observed a lower rate of in-hospital mortality among patients with malnutrition receiving nutritional support compared with those not receiving nutritional support. The magnitude of 21% relative risk reduction was high and comparable to findings from randomized clinical trials from 2019 11  Patients (NOURISH) study 12 evaluated the effects of adding a specialized nutrient-dense oral nutritional supplement compared with placebo supplement among patients with malnutrition who were in the hospital for respiratory or cardiac care. This trial showed a 51% reduction in 90-dayoverall-mortality among treated patients. An updated meta-analysis 23 using individualized patient data reported a 27% relative risk reduction in overall mortality up to 6 months after discharge.
Moreover, our cohort of patients with malnutrition who were not selected differed from those in the clinical studies mentioned above because it included a higher proportion of patients receiving enteral nutrition (4.9% after matching) or parenteral nutrition (2.4% after matching). For example, the EFFORT trial had lower proportions of enteral nutrition (0.8%) and parenteral nutrition (1.2%). 11 Assuming that general health status of patients receiving enteral and parenteral nutrition administration is worse than that of patients not receiving this treatment, our cohort was likely to include more severely ill patients compared with clinical trials completed previously in the non-ICU setting. To further explore a potential association of this discrepancy with results in different study settings, we performed a sensitivity analysis excluding patients receiving enteral or parenteral nutrition. After matching, receiving oral nutritional support alone was associated with a 25% reduction in overall in-hospital mortality risk, which was similar to the findings by Gomes et al. 23 In addition, our study did not assess postdischarge mortality rates, while in Gomes et al, 23 mortality was measured up to 6 months after randomization. The shorter follow-up period in our study may also be associated with differences in survival rates.
As secondary outcomes, we assessed associations with further clinical outcomes beyond the known associations with mortality. We found reduced readmission rates in the group receiving nutritional support. Similarly, the meta-analysis by Gomes et al 23 reported that receiving nutritional support was associated with a reduction in nonelective hospital readmissions, albeit with a large heterogeneity among the included trials. In contrast, the 2 largest randomized clinical trials of nutritional support and mortality completed to our knowledge so far, NOURISH 12 and EFFORT, 11 did not find significant reductions in readmission rates.
Patients in our study receiving nutritional therapy had a reduced risk of discharge to a postacute care facility, although results after accounting for competing risk showed conflicting findings. Three  randomized clinical studies [24][25][26] exploring that end point did not find a difference between intervention and control groups.
To maximize external validity, and thus provide a solid basis to generalize our findings for reallife settings, we included hospitalizations of patients with multiple morbidities, including a broad variety of comorbidities as usually seen in a real-world medical ward population. Our aim was to focus on patients in the hospital with acute illness; therefore, we did not include patients from geriatric units, who often have different characteristics (eg, more admissions to postacute care facilities).
Substantiating our second main finding of this population-based study, the EFFORT trial 11 and the updated meta-analysis 23 found that established or more severe malnutrition, compared with less severe malnutrition or nutritional risk only, was associated with a greater survival benefit. In our cohort, the potential survival benefits associated with a nutritional support were mostly robust across subgroups of age, reason for hospital admission, and comorbidities. However, there was no significant modification associated with the degree of malnutrition. Reductions in risk were greater in patients with lower hospital frailty risk scores and shorter hospital LOS. Whether this finding is directly associated with the nonbeneficial or even unfavorable associations of nutritional support in the critical care setting remains hypothetical. [27][28][29] The main strengths of this study include the use of an unselected population, including patients with a variety of measurable confounding variables, accompanied by strict definitions of nutritional support (ie, high specificity) and patient outcomes. This approach is associated with reduced selection bias and increased external validity. Moreover, the study reflects highly accurate estimates at a national level, incorporating a propensity score-matched analysis of patient-centered hospital care of adult patients with malnutrition as treated in clinical routine, and thus enabling a comparative study setting. Furthermore, consistency of the results with previous findings from randomized clinical trials underlines the generalizability of these data.

Limitations
This study has several limitations. Our data must be interpreted in the context of the study design.
The diagnosis of malnutrition was based on ICD-10 classification, and the procedure of nutritional support was based on CHOP codes, so we could not precisely stratify by the activity or severity of disease or the adequacy of nutrient replacement. As ICD-10 codes for malnutrition are based on the NRS-2002 score, 5 our cohort included not only patients with malnutrition, but also patients at risk for malnutrition. Nonetheless, including a broad spectrum of patients at risk for malnutrition is associated with increased external validity. In addition, a certain risk of misclassification and underreporting needs to be acknowledged, because administrative data were used and codes were provided only starting in 2013. Because of the underestimated prevalence of malnutrition, a 2017 study 30 in Switzerland investigated the diagnostic accuracy of malnutrition codes and found a low sensitivity of 30% but a high specificity of 93.4%. In our study, attribution of the nutritional intervention was at the discretion of the treating physician, and therefore, unmeasured and unmeasurable residual confounding cannot be excluded (eg, expected benefit from nutritional support, previous nutrition therapy, or clinical or laboratory parameters). However, to approximatively address levels of frailty and disease severity, we included in our statistical model aspects of general health status, such as the Charlson Comorbidity Index, the hospital frailty risk score, and the use of palliative care.

Conclusions
In this nationwide population-based cohort of patients with malnutrition in the hospital, receiving nutritional support as prescribed in routine care was associated with decreased in-hospital mortality compared with not receiving nutritional support. These findings were mostly consistent with those of previous large randomized clinical trials and are unlikely to be explained by unmeasured confounding.
JAMA Network Open | Nutrition, Obesity, and Exercise