Functional decline in geriatric rehabilitation ward; is it ascribable to Hospital Acquired Infection? A prospective cohort study.
Categorical variables are described as numbers (%) and were compared by chi-square test or Fisher exact test, as appropriate. Quantitative variables are described as median (interquartile range [IQR]) and were compared by the nonparametric Mann-Whitney test. Characteristics of patients with unavailable ADL data at discharge were compared to those with available data. Considering this latter group, we then compared the groups with and without functional decline in terms of baseline characteristics. Logistic regression modeling was used to estimate odds ratios (ORs) with their 95% confidence intervals (95% CIs). Variables associated with functional decline on univariate analysis at P < 0.15 were then entered into a multivariate logistic regression model. To avoid introducing strongly correlated variables into multivariate models, we assessed correlations by using Cramer’s V for categorical variables and the nonparametric Spearman’s rank correlation coefficient (Rho) for quantitative variables. All models including albumin level were systematically adjusted for C-reactive protein (CRP) level, as appropriate (35). Finally, and in accordance with our hypothesis, we examined whether HAI occurrence potentially mediated the relation between comorbidities and functional decline, as illustrated in the conceptual framework shown in Fig. 1. According to Baron and Kenny (38), evidence for a partial mediating effect was assessed by the statistical significance (38–41) of the following associations:
1) between comorbidities as the independent exposure of the interest (A) and functional decline as the outcome (Y),
2) between comorbidities and HAI as the mediating factor (M),
3) and between HAI and functional decline
and by a reduced effect of comorbidities on functional decline after adjusting for HAI.
To test the robustness of our results, we performed three sensitivity analyses on the final models. Using the hypothesis of maximal bias, we first considered that all patients with missing discharged ADL data had functional decline, and second that these patients had no functional decline. Finally, we used a multiple imputation approach with the multiple-multivariate imputation-by-chained-equations procedure with the missing-at-random assumption. We used all baseline covariates and outcomes together to impute missing data values and independently analyzed 20 copies of the data.
All tests were two-tailed. P ≤ 0.05 was considered statistically significant. Data were analyzed by using STATA v11.0 (StataCorp, College Station, TX, USA).