Predictors of 30-day readmissions and Model Discrimination
The 30-day readmission rate was 8.2% (n=243). By univariate analysis,
patients who were readmitted had a higher prevalence of CHF (36.6% vs
26.2%, p<0.001), ischemic heart disease (52.3% vs 43.2%,
p=0.006), atherosclerosis (5.3% vs 1.9%, p<0.001) NSTEMI
(1.6% vs 0.2%,
p=0.006), anemia (20.6% vs 10.9%, p<0.001), obesity (13.6%
vs 9.3%, p=0.028), acute kidney failure (4.9% vs 2.0% p=0.003), end
stage renal disease (5.8% vs 2.2%, p<0.001), chronic kidney
disease (26.7% vs 15.1%, p<0.001), and chronic obstructive
pulmonary disease (26.3% vs 11.6%, p<0.001) noted during
their index admission (Table 1). Of the Elixhauser Comorbidity Index
variables, readmitted patients had statistically significant higher
prevalence of CHF, peripheral vascular disorders (23.9% vs 16.3%,
p=0.002), complicated hypertension (34.6% vs 20.8%,
p<0.001), chronic pulmonary disease, complicated diabetes
(13.6% vs 6.7%, p<0.001),
renal failure, obesity, fluid and electrolyte disorders (8.6% vs 3.6%,
p<0.001), anemia, and drug abuse (1.2% vs 0.2%, p=0.031). In
contrast, patients who were not readmitted had a statistically
significant higher prevalence of uncomplicated hypertension (62.8% vs
52.3%, p=0.001). After multiple regression analysis, non-routine
disposition (HR 2.19 [1.5-3.21], p<0.0001), such as
discharge to Home Health Care (HHC) or a Skilled Nursing Facility
(SNF), chronic pulmonary disease (HR 1.76 [1.3-3.29],
p=0.0003), renal disease (HR 1.96 [1.44-2.67]), p<0.0001),
and anemia (HR 1.55 [1.12-2.17], p=0.009) were significant
predictors of readmission in this cohort (Figure 2). Peripheral vascular
disease approached significance in this analysis (HR 1.39 [0.98-
1.98], p=0.0686). Notable non-predictors included age, sex, diabetes,
cardiomyopathy and HTN. Of the available peri-procedural complications,
including pericardial effusion, cardiac tamponade, cerebral infarction,
device thrombosis, venous or arterial thrombosis, intra- or post-
procedure cardiac arrest, arteriovenous fistula formation, and acute
renal failure, only acute renal failure (p = 0.003) and venous
thromboembolism (p = 0.002) were predictive of readmission in a
statistically significant fashion (Table 3). Cerebral infarction
approached significance but was ultimately found to be non-significant
(p = 0.054). Overall, the C-statistic of the model was 0.70.
Figure 2. Multivariable model of 30-day readmissions in
patients undergoing LAAC procedures. Parentheses indicate 95%
confidence intervals.