Statistical analysis:
Categorical outcomes were analyzed by chi-square test with continuity
correction or Fisher’s exact test, wherever one or more expected cell
size was less than 5. Numerical variables were tested for normality
using the test for normality (Kolmogorov–Smirnov). Normally distributed
variables were compared by unpaired t-test after evaluating equality of
variance by Levene’s test (F test), whereas skewed variables were
analyzed with the Mann–Whitney U-test. Univariate analysis was done to
identify predictors of long stay. Logistic regression with backward
stepwise elimination was done to identify independent predictors of long
stay and a p-value of less than .05 was taken as significant. A weighted
prediction score was devised using beta coefficients of independent
variables derived from the logistic model (multiplied by 10 and rounded
off to nearest integer)16. The discriminatory power
and calibration of the prediction score was evaluated using area under
Receiver Operating Characteristic (ROC) curve. The optimal cut-off for
the weighted score was calculated using the ROC curve and estimated
sensitivity, specificity and positive predictive value of the cut
off17. The analysis was done using statistical
software packages IBM-SPSS v.25 (SPSS Inc, NY, USA).