Nutritional risk screening score as an independent predictor of nonventilator hospital-acquired pneumonia: a cohort study of 67,280 patients

We conducted a retrospective cohort study, including all inpatients admitted between September 1, 2017, through June 30, 2020, at Wenzhou People’s Hospital (a 1500-bed tertiary teaching hospital in Zhejiang, China). Patients who were pregnant, younger than 18 years of age or greater than 90 years of age, length of hospital stay < 48 h, received mechanical ventilation during hospitalization, and lack of nutritional risk screening score were excluded from the analysis. If patients were readmitted during the study period, only their first admission was considered. The study was approved by the Ethics Review Committee of Wenzhou people’s Hospital [approval no. WRY2018070]. Given the retrospective nature of the study, the requirement of informed consent was waived. This paper was reported in line with the STROBE guidelines [20].

All adult patients, except pregnant women, underwent nutritional risk screening. The nutritional risk screening was performed within 24 h after admission by ward nursing staff who were trained to conduct using the NRS-2002 tool.

The NV-HAP data was obtained from the Xinglin system [21]. This system is a web-based, real-time monitoring system of nosocomial infection, which automatically identify symptoms of infections and clinical data such as fever, positive bacterial culture, and elevated inflammatory response markers for initial diagnoses. Meanwhile, the system is also used to transfer nosocomial infection cases identified by the clinicians to senior infection control practitioner for a definitive diagnosis. In case of a disagreement between the two sides, a consensus was made via discussions. Nonventilator hospital-acquired pneumonia (NV-HAP) is defined as a pneumonia not present or incubating at the time of hospital admission and occurring at least 48 h after admission in patients not receiving invasive mechanical ventilation during hospitalization [22]. The diagnostic criteria used in the present study for NV-HAP strictly adhered to the 2018 version of the Chinese guidelines [22]. The 2018 version of the Chinese guidelines is compatible with the guidelines issued by the American Thoracic Society [23].

Admission data collected from the electronic medical record system included age; sex; drinking status; smoking status, comorbidities, admission category, Barthel Index, Morse Fall Scale, and season of admission. The Barthel Index (BI) [24] and the Morse Fall Scale [25] were used to assess the patient’s level of independence and nursing-related complications, respectively. Charlson comorbidity index (CCI) was used to measure the burden of comorbid conditions [26].

Based on the outcome and exposure to the hospital environment, we added a covariate termed “time at risk” into the model. For NV-HAP patients, “time at risk” was calculated as the number of days between the admission day and date of diagnosis of NV-HAP. For non-NV-HAP patients, the “time at risk” corresponded to the total hospital days. We collected information from the Xinglin system concerning clinical procedures (including a central venous catheter, indwelling urinary catheter, surgery, parenteral nutrition, and enteral tube feeding), other nosocomial infections, and the use of specific classes of medications such as antacids, sedatives, nonsteroidal anti-inflammatory drug (NSAID), systemic steroid, inhaled steroid, and anticoagulant during the “time at risk” period.

Non-normal continuous variables were presented as medians (Q1-Q3) and compared using Mann-Whitney U test. Categorical variables were presented as numbers (proportion) and compared using Chi-square test or Fisher’s exact test. To avoid bias caused by missing NRS score data, the characteristics of individuals with missing data were compared with those with complete data. As less than 1% of the covariates were missing, the missing data were not dealt with. Logistic regression analyses were used to estimate odds ratios (ORs) and 95% confidence intervals (95% CIs) for the association between NRS score and risk of NV-HAP. Firstly, possible collinearity was assessed based on the variance inflation factor (VIF); variables with VIF > 10 were removed from the Model. Secondly, we used four different logistic regression models to examine the associations of nutritional risk screening score and the risk for NV-HAP. The Non-adjusted Model examined the association between NRS score and NV-HAP without adjustment for any covariates. Model I included demographic characteristics (age and sex). Model II made an additional adjustment for variables that, when added to this model, changed in effect estimate of more than 10% [27], included the covariates in Model I plus stroke, Charlson comorbidity index, time of risk, central venous catheter, enteral tube feeding, Barthel Index, Morse Fall Scale. The association of each covariate with NV-HAP is shown in Supplementary Table S2. Model III (the fully adjusted model) included the covariates in Model II plus the other covariates listed in Table 1. Thirdly, these analyses were performed on unique patients, making it possible for a patient with multiple admissions; therefore, risk estimation was also performed using the generalized estimation equation (GEE) method with a logit link and exchangeable correlation matrix while adjusting for the possible dependence in the outcome introduced by repeated admissions. Finally, to assess the homogeneity of effects, subgroup analyses and interaction tests were performed for the covariates shown in Table 1.

Statistical analyses were performed with the R software (version 3.4.3; http://​www.​R-project.​org) and EmpowerStats software (www.​empowerstats.​com, X&Y solutions, Inc. Boston MA). A two-tailed P-value of ≤0.05 was considered to be statistically significant.

There were 154,024 admissions to the medical centre from September 1, 2017, through June 30, 2020. After excluding admissions with younger than 18 years of age or greater than 90 years of age (n = 13,491), length of hospital stay < 48 h (n = 9207), received mechanical ventilation during hospitalization (n = 1658), pregnancy (n = 37,891), lack of nutritional risk screening score (n = 173), and repeated admissions (n = 24,324), a total of 67,280 unique patients were included in the final analysis (Fig. 1). Baseline characteristics of study participants by NRS score are listed in Table 1. In the present study, 4578 (6.8%) patients were at nutritional risk (NRS-2002 ≥ 3). There were significant differences in baseline characteristics between patients with NRS score < 3 and those with NRS score ≥ 3 (Table 1).

The incidence of NV-HAP in the cohort for the NRS < 3 group and NRS score ≥ 3 group was 0.4% (232/62702) and 2.6% (121/4578), respectively (Table 1). The proportion of patients with NV-HAP was significantly higher in the NRS ≥3 groups (Fig. 2a). The incidence of NV-HAP showed an NRS score-dependent increase (P for trend< 0.001). The incidence of NV-HAP was 0.2, 0.8, 1.1, 2.3, 2.5, 4.7, and 15.1% for NRS scores of 0, 1, 2, 3, 4,5, and ≥ 6, respectively (Fig. 2b).

Results of VIF analysis for variables showed that there was no collinearity bias (Table S1 in the Supplementary Appendix). The unadjusted and multivariate-adjusted analyses of the relationship between NRS score and NV-HAP are shown in Table 2. NRS score, whether considered a categorical or continuous variable, was independently associated with the risk of NV-HAP in different multivariate logistic regression models. Patients with NRS score ≥ 3 were at a higher risk of NV-HAP (OR = 7.31; 95%CI: 5.86, 9.31) than those with NRS score was < 3. After adjusting for age and sex (Model I), the association remained unchanged (OR = 4.39;95% CI: 3.47, 5.55).

After additional adjustment for stroke, CCI, time of risk, central venous catheter, enteral tube feeding, Barthel Index, and Morse Fall Scale (Model II), the association did not change (OR = 1.97;95% CI: 1.52, 2.56). Finally, in the fully adjusted model (Model III), the OR was 2.06 (95% CI: 1.58, 2.70). Similarly, per 1-point increase in the NRS score was associated with a 105,74,28, and 30% higher risk of NV-HAP in different multivariate logistic regression models.

A total of 174 patients with missing NRS score data were found. After excluding 40 patients with repeated admissions, 134 patients with missing NRS score data were retained, accounting for 0.2% of all analysed patients (134/ 67,280). There was no significant difference in most baseline characteristics, including the Charlson comorbidity index, between the two groups (Table S3 in the Supplementary Appendix).

In the sensitivity analyses, similar results were observed in analyses using a GEE method with a logit link and exchangeable correlation matrix (Table S4 in the Supplementary Appendix).

We further performed stratified and interaction analyses to assess the effect of the NRS score (< 3/≥3) on NV-HAP in various subgroups (Fig. 3). In the subgroup analyses, the NRS score was associated with a greater risk of NV-HAP in most subgroups. The difference was not statistically significant in some subgroups, probably due to the limited sample size. Tests for interaction showed no statistically significant differences for effect modification by other covariates (all Pvalue for interaction > 0.05).