Comparison of clinical characteristics of community-acquired APN
A total of 241 patients with community-acquired APN caused by Enterobacteriales were recruited. Of these, E. coli was the most common pathogen, accounting for 90% of Enterobacteriales. The proportions of the ESBL+ and CIP-R groups were 31.1% (75/241) and 36.1% (87/241), respectively. There was no difference in the composition of uropathogens between the ESBL+ and ESBL- groups, as well as between the CIP-R and CIP-S groups (Table 1).
Table 2 shows clinical characteristics of patients with community-acquired APN. The mean age was not significantly different between the ESBL+ and ESBL- groups. The ESBL+ group showed higher Charlson’s comorbidity index (2.24 ± 2.35 vs. 1.55 ± 1.74, p = 0.026) and higher proportion of having past history of admission during 1 year prior to inclusion (41.3% vs. 23.5%, p = 0.008) than the ESBL- group, but there were no significant differences in the clinical features between the two groups.
In comparison, the mean age was higher in the CIP-R group than in the CIP-S group (68.67 ± 15.61 vs. 63.32 ± 17.14, p = 0.014). Charlson’s comorbidity index was similar between the CIP-S and CIP-R groups, while the proportion of patients who had a history of admission during 1 year prior to inclusion was higher in the CIP-R group than in the CIP-S group (44.8% vs. 20.1%, p < 0.001). In addition, the proportion of patients with a history of antibiotic usage during 1 year prior to inclusion was higher in the CIP-R group than in the CIP-S group (47.1% vs. 26.6%, p = 0.002).
As for the treatment, the probability of discordance between the initial antimicrobial regimen and the antimicrobial susceptibility of causative organisms was higher in the ESBL+ group than in the ESBL- group (53.33% vs. 0.61%, p < 0.001) and in the CIP-R group than in the CIP-S group (29.9% vs. 9.9%, p < 0.001).
Comparison of outcomes of community-acquired APN
Table 3 shows outcomes of community-acquired APN. A patient in the ESBL+ group incurred higher medical costs compared to a patient in the ESBL- group (USD 3,730.2 vs. 3,119.3, p < 0.001). In detail, hospitalization expenditure (USD 1,331.2 vs. 1,099.0, p = 0.018), meal (USD 137.3 vs. 107.2, p = 0.008), cost of medication (USD 505.6 vs. 334.6, p < 0.001), and procedure or operation charge (USD 376.5 vs. 271.2, p = 0.018) were higher in the ESBL+ group than in the ESBL- group. Similarly, total medical costs were higher in the CIP-R group compared to that of the CIP-S group (USD 3,730.2 vs. 3,119.3, p = 0.005). In detail, consultation fee (USD 141.7 vs. 113.0, p = 0.005), hospitalization expenditures (USD 1,360.7 vs. 1,067.5, p = 0.002), meals (USD 145.9 vs. 103.2, p = 0.005), cost of medication (495.6 vs. 346.3, p = 0.005), and procedure or operation charges (USD 326.2 vs. 275.0, p = 0.045) were higher in the CIP-R group than in the CIP-S group.
The total length of hospital stay was longer in the ESBL+ group than in the ESBL- group (11 vs. 8 days, p < 0.001), as well as in the CIP-R group than in the CIP-S group (11 vs. 8 days, p < 0.001). No significant differences were observed in the proportion of clinical failure and change in Braden scale between the ESBL+ and ESBL- groups and the CIP-R and CIP-S groups, respectively.
Risk factors for higher medical costs and longer length of stay using the log-linear regression model
The results in Tables 4 and 5 display the recovered average marginal effect of each statistically significant variable on total medical costs and length of stay, respectively. Note that the coefficients are the estimates obtained from the log-linear regression model (i.e., the outcome variable is ln(total cost) and ln(length of stay)). Therefore, the coefficient estimates should not be interpreted as a marginal effect on the total costs or length of stay but as semi-elasticity. In the last column, we report the recovered average marginal effect of a unit increase in a control variable on the level outcome (i.e., total costs and length of stay), adjusting for all other factors (ceteris paribus).
Holding all other factors constant, the treatment cost of APN for a case in which ESBL-producing Enterobacteriales is a causative pathogen would be, on average, 27%, or USD 1,211 (p = 0.026) more expensive than non-ESBL-producing pathogen. A patient who is a year older would incur USD 23 (p = 0.040) more, having any structural problems in the urinary tract would incur USD 1,231 (p = 0.015) more, and a unit increase in the Pitt bacteremia score would incur USD 767 (p <0.001) higher costs, all other things constant. However, the total costs were lower if one had had a history of antibiotic usage during 1 year prior to inclusion (USD 987 lower; p = 0.017) or had a history of urinary tract operation during the 3 months prior to inclusion (USD 1,704 lower, p = 0.04). A point higher on the Braden scale at admission was associated with lowering of costs by USD 240 (p <0.001).
Adjusting for other factors, having a case in which ESBL-producing Enterobacteriales is a causative pathogen would explain staying 22% longer or 2 more days (p = 0.050) in the hospital than non-ESBL-producing Enterobacteriales. A patient who is 10 years older would, on average, would have to stay for half a day longer (p = 0.045). Any structural problems in the urinary tract explain a longer stay (2.4 days longer; p = 0.032), and moving from 0 to 5 on the Pitt bacteremia score would explain four more days (p = 0.038) in the hospital.
Risk factors for higher medical costs and longer length of stay using the quantile regression model
Figure 1 and 2 present the point estimates and 95% confidence intervals (bands) from quantile regression. The corresponding point estimates and standard errors can be found in Supplementary Table 1 and 2, respectively, in which we report the conditional quantile at 0.1 through 0.9 (0.1 quantile and 10th percentile are synonymous). Note that the outcome is in log-scale; thus, the coefficient estimate can be interpreted as semi-elasticities. We do not recover the average marginal effect because the semi-elasticity serves the purpose of presenting the impact of regressors on the conditional distribution of the outcomes. For example, if the coefficient estimate is 0.02, then a one-unit change in that variable is associated with a proportionate increase of 0.02 or 2%.
Figure 1 presents the quantile regression results for the total costs (in log scale). This shows that age has a positive relationship with incurring higher total costs at the 50th or 60th percentile but not at the tails. That is, at the median or at the 60th percentile, being 1 year older is associated with 0.8% (p = 0.030) higher treatment expenditures. In the higher quantiles, the point estimates suggest that having any structural problems in the urinary tract is associated with 31‒50% higher medical costs. Although having ESBL-producing Enterobacteriales as a causative pathogen is, on average, associated with 27% higher medical costs (Table 4), the quantile regression results suggest that the main impact was made in the 60th percentile, which incurred 27% higher costs. Similarly, this suggests that history of urinary tract operation during the 3 months prior to inclusion is associated with significantly lower (50‒116%) medical costs in the higher percentiles. The initial Braden scale at the time of admission or the Pitt bacteremia score seemed to affect the outcome distribution mostly throughout the quantiles.
Figure 2 presents the quantile regression results for the length of stay (in log scale). While having any structural problem in the urinary tract was associated with a 22% longer stay in the hospital on average (Table 5), and the impact was significant in the upper quantiles (32‒36% longer in the 70th and 80th percentiles). History of having a urinary tract operation during the 3 months prior to inclusion was associated with a shorter length of stay in the upper quantiles (54‒78% shorter), but with a longer length of stay in the lowest quantile (65% longer). A one-unit increase in the Braden scale at the time of admission explains a 4.3% shorter stay at the 70th percentile.