Predictive Models for Identification of Hospitalized Patients Harboring KPC-Producing Klebsiella pneumoniae

Abstract

The production of Klebsiella pneumoniae carbapenemases (KPCs) by Enterobacteriaceae has become a significant problem in recent years. To identify factors that could predict isolation of KPC-producing K. pneumoniae (KPCKP) in clinical samples from hospitalized patients, we conducted a retrospective, matched (1:2) case-control study in five large Italian hospitals. The case cohort consisted of adult inpatients whose hospital stay included at least one documented isolation of a KPCKP strain from a clinical specimen. For each case enrolled, we randomly selected two matched controls with no KPCKP-positive cultures of any type during their hospitalization. Matching involved hospital, ward, and month/year of admission, as well as time at risk for KPCKP isolation. A subgroup analysis was also carried out to identify risk factors specifically associated with true KPCKP infection. During the study period, KPCKP was isolated from clinical samples of 657 patients; 426 of these cases appeared to be true infections. Independent predictors of KPCKP isolation were recent admission to an intensive care unit (ICU), indwelling urinary catheter, central venous catheter (CVC), and/or surgical drain, ≥2 recent hospitalizations, hematological cancer, and recent fluoroquinolone and/or carbapenem therapy. A Charlson index of ≥3, indwelling CVC, recent surgery, neutropenia, ≥2 recent hospitalizations, and recent fluoroquinolone and/or carbapenem therapy were independent risk factors for KPCKP infection. Models developed to predict KPCKP isolation and KPCKP infection displayed good predictive power, with the areas under the receiver-operating characteristic curves of 0.82 (95% confidence interval [CI], 0.80 to 0.84) and 0.82 (95% CI, 0.80 to 0.85), respectively. This study provides novel information which might be useful for the clinical management of patients harboring KPCKP and for controlling the spread of this organism.

INTRODUCTION

Since 2010, Klebsiella pneumoniae carbapenemase (KPC)-producing Enterobacteriaceae have caused numerous outbreaks of severe nosocomial infections in the northeastern United States, Israel, Greece, and Italy, and they are now considered endemic in these areas (1–9). Not only do KPC-producing organisms hydrolyze carbapenems, but also, they are often resistant to multiple other antibiotics (7, 9–11). Consequently, they are invariably associated with high treatment failure rates (3, 4, 7).

Several relatively small studies have investigated risk factors for infection and/or colonization by carbapenem-resistant isolates of K. pneumoniae and other Gram-negative bacteria (5, 8, 12–16). A recent review of these efforts found that the factors associated with colonization, infection, and mortality caused by these organisms vary considerably with the species and enzymatic profile of the isolate (5). Less attention has been focused specifically on KPC-producing strains of K. pneumoniae (KPCKP), and these studies have also been fairly small (17–20). The reported rates of mortality related to KPCKP infections vary widely, from 22% to 72% (3, 4, 7, 21). The risk obviously depends on factors like population age and underlying disease/comorbidity profiles, which were highly heterogeneous in the studies cited above. The criteria used to distinguish true KPCKP infections and simple colonization can also influence mortality rates. The term “infection” is frequently applied to cases characterized by isolation of KPCKP from clinical specimens (rather than rectal swabs collected for screening purposes). However, the presence of KPCKP in such samples—especially sputum and urine—can also reflect simple colonization. Therefore, this definition can favor overtreatment and distorts assessments of clinical outcomes.

To clarify the inpatient risk factors associated with isolation of a KPCKP strain, we conducted a large, retrospective study in five Italian hospitals, where these organisms are an increasingly frequent cause of serious disease. A subgroup analysis was also carried out to identify risk factors specifically associated with true KPCKP infection.

MATERIALS AND METHODS

Setting, study design, and patients.

A matched case-control analysis was conducted to identify risk factors associated with KPCKP isolation. Cases and controls were identified via databases maintained by the microbiology laboratories of five large, full-service teaching hospitals in Italy. During the study period (1 January 2010 to 31 December 2012), rectal swab screening was not a routine admission procedure in any of the hospitals (although it was used regularly in certain high-risk settings, such as intensive care units [ICUs]). The case cohort thus consisted of adult inpatients consecutively admitted to participating hospitals within the study period, whose hospital stay included at least one documented isolation of a KPCKP strain from a clinical specimen (excluding stool and/or rectal swabs). Each patient was included only once, at the time of the first KPCKP isolation (index culture). For each case enrolled, we randomly selected two matched controls with no KPCKP-positive cultures of any type during their hospitalization. Matching involved hospital, ward, and month/year of admission, as well as time at risk for KPCKP isolation (defined below).

The initial analysis involving all members of the case and control cohorts was conducted to identify risk factors associated with recovery of a KPCKP strain from a clinical specimen. Each case was also classified as KPCKP colonization or KPCKP infection, and potential risk factors for the latter were explored in a separate case-control subanalysis (i.e., infected cases versus their controls). Two infectious-disease specialists independently reviewed data available at the time of each case patient's discharge or death. Cases were considered infections according to standard definitions (22) and if they had received an antibiotic regimen consistent with the KPCKP isolate antibiogram. If either of the criteria was not met, the case was classified as colonization and excluded from the subgroup analysis. When judgments were discordant, the reviewers reexamined the data jointly and reached a consensus decision. When data necessary for classification were unavailable, the case was excluded from the subgroup analysis.

Microbiological procedures.

The Vitek 2 system (bioMérieux, Marcy l'Etoile, France) was used in all participating centers for isolate identification and antimicrobial susceptibility testing. Vitek MICs were classified according to Clinical and Laboratory Standards Institute breakpoints (23), except those for tigecycline and colistin, which were confirmed by Etest (bioMérieux) and classified according to EUCAST breakpoints (24). bla genes were identified by PCR and sequencing, as previously described (25, 26).

Data collected and variables analyzed.

Medical records and computerized hospital databases were reviewed to identify patient demographics; hospital admission characteristics, including date, source (i.e., home versus another health care facility), and admitting ward; and vital status at discharge. For cases, time at risk for KPCKP isolation was defined as the interval from admission (to the study facility or, in cases of transfer, admission to the transferring health care facility) to index culture (see above); controls were matched for time at risk with cases based on the their entire hospital stay. KPCKP isolates were considered hospital acquired if the index culture had been collected >48 h after admission and signs and symptoms of infection had been absent at admission. If cultures had been collected ≤48 h after the admission date, the isolate was classified as “health care associated” or “community acquired” (27).

The variables investigated as risk factors for KPCKP isolation included underlying diseases and comorbidities analyzed singly and collectively on the basis of the Charlson index (28), indwelling invasive devices at index culture, and aspects of the recent medical history (contact with the health care system, infectious episodes, invasive procedures, drug therapy, other treatments). Table 1 shows specific assessment time frames for each variable.

TABLE 1.

Univariate analysis of risk factors for KPCKP strain isolation and for KPCKP infection^a

Variable	Isolation (n = 657)	Controls (n = 1,314)	OR (95% CI)	P	Infection (n = 426)	Controls (n = 852)	OR (95% CI)	P
Patient characteristics
Age (yr)	70 (57–78)	66 (49–77)	-	<0.001	68.5 (56–78)	66 (50–77)	-	0.02
Male sex	372 (56.6)	708 (53.8)	1.11(0.92–1.35)	0.24	238 (55.9)	462 (54.2)	1.07 (0.84–1.36)	0.58
Transfer from another health care facilityb	22 (3.3)	14 (1.1)	3.21 (1.56–6.84)	<0.001	10 (2.3)	11 (1.3)	1.84 (0.69–4.81)	0.16
Bedridden	241 (36.7)	352 (26.8)	1.58 (1.38–1.94)	<0.001	146 (34.3)	219 (25.7)	1.51 (1.16–1.95)	0.001
Comorbidityc
COPD	123 (18.7)	163 (12.4)	1.62 (1.24–2.11)	<0.001	66 (15.5)	114 (13.4)	1.19 (0.84–1.67)	0.31
Cardiovascular disease	272 (41.4)	439 (33.4)	1.40(1.15–1.71)	<0.001	167 (39.2)	295 (34.6)	1.22 (0.95–1.56)	0.11
Cerebrovascular disease or dementia	133 (20.2)	153 (11.6)	1.92(1.48–2.50)	<0.001	61 (14.3)	103 (12.1)	1.21 (0.85–1.73)	0.26
Cancer, solid tumor	167 (25.4)	279 (21.2)	1.26 (1.01–1.58)	0.03	103 (24.2)	195 (22.9)	1.07 (0.81–1.42)	0.61
Cancer, hematologic	77 (11.7)	95 (7.2)	1.70 (1.22–2.36)	<0.001	53 (12.4)	72 (8.4)	1.54 (1.03–2.28)	0.02
Diabetes mellitus	152 (23.1)	209 (15.9)	1.59 (1.24–2.02)	<0.001	105 (24.6)	142 (16.7)	1.63 (1.22–2.19)	<0.001
Chronic renal failure	109 (16.6)	113 (8.6)	2.11 (1.57–2.82)	<0.001	69 (16.2)	80 (9.4)	1.86 (1.30–2.67)	<0.001
Liver disease	120 (18.6)	128 (9.7)	2.07 (1.56–2.73)	<0.001	93 (21.8)	88 (10.3)	2.42 (1.73–3.37)	<0.001
HIV infection	7 (1.1)	23 (1.7)	0.60 (0.21–1.46)	0.24	6 (1.4)	14 (1.6)	0.85 (0.27–2.39)	0.75
Solid organ transplantation	37 (5.6)	34 (2.6)	2.24 (1.35–3.72)	<0.001	25 (5.9)	25 (2.9)	2.06 (1.12–3.79)	0.01
Charlson index ≥3	331 (50.4)	149 (11.3)	7.93 (6.27–10.05)	<0.001	226 (53.1)	126 (14.8)	6.51 (4.93–8.59)	<0.001
Neutropenia (ANC < 500 cells/mm3)	30 (4.6)	25 (1.9)	2.46 (1.38–4.41)	<0.001	25 (5.9)	18 (2.1)	2.89 (1.49–5.68)	<0.001
Presence of invasive devicesc
CVC	373 (56.7)	374 (28.4)	3.30 (2.70–4.03)	<0.001	247 (58.0)	239 (28.0)	3.54 (2.75–4.55)	<0.001
Urinary catheter	438 (66.6)	361 (27.4)	5.27 (4.29–6.49)	<0.001	272 (63.8)	226 (26.5)	4.89 (3.78–6.33)	<0.001
Nasogastric tube	203 (30.9)	182 (13.8)	2.78 (2.19–3.51)	<0.001	130 (30.5)	114 (13.4)	2.84 (2.11–3.82)	<0.001
Surgical drain	161 (24.5)	123 (9.3)	3.14 (2.41–4.10)	<0.001	103 (24.2)	77 (9.0)	3.21 (2.29–4.49)	<0.001
Biliary stent	11 (1.6)	22 (1.6)	1.00 (0.43–2.16)	1.00	5 (1.2)	15 (1.8)	0.66 (0.19–1.93)	0.42
PEG	19 (2.9)	18 (1.4)	2.14 (1.05–4.36)	0.01	11 (2.6)	15 (1.8)	1.48 (0.61–3.48)	0.33
Recent contact with health care system
Regular outpatient/day-hospital cared	148 (22.5)	214 (16.3)	1.49 (1.17–1.90)	<0.001	95 (22.3)	153 (18.0)	1.31 (0.97–1.76)	0.06
Acute-care hospitalization(s)e	425 (64.7)	385 (29.3)	4.42 (3.60–5.42)	<0.001	265 (62.2)	368 (43.2)	2.16 (1.69–2.77)	<0.001
1	186 (28.3)	279 (21.2)	1.46(1.17–1.82)	<0.001	115 (27.0)	277 (32.5)	0.77 (0.59–1.00)	0.04
≥2	239 (36.3)	114 (8.6)	6.01 (4.65–7.79)	<0.001	150 (35.2)	99 (11.6)	4.13 (3.06–5.57)	<0.001
≥3	132 (20.1)	70 (5.3)	4.46 (3.25–6.16)	<0.001	73 (17.1)	55 (6.4)	2.99 (2.03–4.42)	<0.001
≥4	77 (11.7)	41 (3.1)	4.12 (2.74–6.25)	<0.001	38 (8.9)	27 (3.1)	2.99 (1.74–5.17)	<0.001
Hospitalization with ICU care	195 (29.6)	47 (3.6)	11.37 (8.06–16.26)	<0.001	120 (28.2)	38 (4.5)	8.40 (5.63–12.71)	<0.001
Recent infectious episodes
Urinary tract infectionf	128 (19.5)	76 (5.8)	5.94 (2.88–5.40)	<0.001	68 (16.0)	47 (5.5)	3.25 (2.16–4.92)	<0.001
Other bacterial infectionsf	286 (43.5)	150 (11.4)	5.98 (4.72–7.57)	<0.001	190 (44.6)	103 (12.1)	5.85 (4.38–7.83)	<0.001
Isolation of ESBL-producing bacteriad	75 (11.4)	30 (2.3)	5.51 (3.51–8.82)	<0.001	46 (10.8)	14 (1.6)	7.24 (3.85–14.43)	<0.001
Isolation of MRSAd	36 (5.5)	13 (0.9)	5.80 (2.97–11.99)	<0.001	26 (6.1)	9 (1.1)	6.09 (2.73–14.88)	<0.001
Recent invasive procedure(s)f
Surgical procedures	285 (43.3)	315 (23.9)	2.42 (1.97–2.98)	<0.001	180 (42.2)	207 (24.3)	2.28 (1.76–2.94)	<0.001
Hemodialysis	51 (7.7)	25 (1.9)	4.33 (2.60–7.37)	<0.001	37 (8.7)	17 (2.0)	4.67 (2.52–8.95)	<0.001
Endoscopy	107 (16.3)	86 (6.5)	2.77 (2.03–3.80)	<0.001	68 (16.0)	55 (6.5)	2.75 (1.86–4.09)	<0.001
Mechanical ventilation	145 (22.1)	186 (14.2)	1.71 (1.33–2.20)	<0.001	103 (24.2)	117 (13.7)	2.00 (1.47–2.72)	<0.001
Recent therapyf
Radiotherapy	37 (5.6)	42 (3.2)	1.80 (1.11–2.91)	0.009	23 (5.4)	28 (3.3)	1.67 (0.91–3.06)	0.06
Chemotherapy	160 (24.5)	274 (20.8)	1.22 (0.97–1.53)	0.07	132 (30.9)	193 (22.6)	1.53 (1.17–2.01)	0.001
Immunosuppressants	49 (7.4)	62 (4.7)	1.62(1.08–2.43)	0.01	36 (8.4)	39 (4.6)	1.92 (1.17–3.16)	0.005
Corticosteroids	178 (27.1)	188 (14.3)	2.22 (1.75–2.82)	<0.001	106 (24.9)	129 (15.1)	1.86 (1.37–2.50)	<0.001
Antimicrobial therapy, any type	518 (78.8)	485 (36.9)	6.36 (5.09–7.98)	<0.001	332 (77.9)	315 (37.0)	6.02 (4.57–7.96)	<0.001
Aminoglycosides	39 (5.9)	34 (2.6)	2.37(1.44–3.91)	<0.001	26 (6.1)	21 (2.5)	2.57 (1.37–4.87)	0.001
β-Lactam–β-lactamase inhibitors	310 (47.2)	266 (20.2)	3.51 (2.85–4.33)	<0.001	212 (49.8)	167 (19.6)	4.06 (3.12–5.28)	<0.001
Fluoroquinolones	249 (37.9)	217 (16.5)	3.08 (2.47–3.84)	<0.001	164 (38.5)	148 (17.4)	2.98 (2.27–3.91)	<0.001
Oxyiminocephalosporins	91 (13.8)	161 (12.2)	1.15 (0.86–1.52)	0.31	57 (13.4)	101 (11.8)	1.15 (0.79–1.65)	0.43
Carbapenems	227 (34.5)	115 (8.7)	5.50 (4.25–7.13)	<0.001	142 (33.3)	80 (9.4)	4.82 (3.51–6.64)	<0.001
Glycopeptides	186 (28.3)	173 (13.1)	2.60 (2.04–3.31)	<0.001	109 (25.6)	119 (14.0)	2.12 (1.56–2.86)	<0.001
Others	239 (36.3)	140 (10.6)	4.79 (3.75–6.12)	<0.001	162 (38.0)	91 (10.7)	5.13 (3.79–6.95)	<0.001
≥3 different antimicrobials	251 (38.2)	111 (8.4)	6.70 (5.18–8.68)	<0.001	163 (38.3)	74 (8.7)	6.51 (4.73–8.99)	<0.001

^aResults are shown as number (%) or median (range). ANC, absolute neutrophil count; CI, confidence interval; COPD, chronic obstructive pulmonary disease; CVC, central venous catheter; HIV, human immunodeficiency virus; ESBL, extended-spectrum beta-lactamase; MRSA, methicillin-resistant Staphylococcus aureus; OR, odds ratio; PEG, percutaneous endoscopic gastrostomy.

^bIncluding acute-care and long-term-care facilities.

^cAt the time of the index culture (cases) or at any time during hospitalization (controls).

^dWithin the 3 months preceding index culture (cases) or at any time during hospitalization (controls).

^eWithin the 12 months preceding index hospital admission for cases and controls.

^fZero to thirty days before the index culture (cases) or at any time during hospitalization (controls).

Statistical analysis.

Continuous variables are reported as means ± standard deviations (SD) or medians (interquartile range [IQR]); numbers and percentages are reported for categorical variables. The Student t and Mann-Whitney U tests were used to compare normally and nonnormally distributed continuous variables, respectively. Categorical variables were evaluated with the χ² or two-tailed Fisher's exact test. Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated for all associations.

Multivariate logistic regression models were used to adjust for potential confounding. Variables associated with KPCKP isolation or KPCKP infection in univariate analysis (P ≤ 0.10) were included in a logistic regression analysis, and a backward stepwise approach was used to identify those independently associated with each outcome (P ≤ 0.05). These factors were included in the corresponding predictive model (one for isolation and one for infection). The power of each model to discriminate between cases and controls was expressed as the area under the receiver-operating characteristic curve (AUROC). For each prediction rule, sensitivity and specificity were calculated for different cutoff values. Statistical analyses were performed with the Intercooled Stata program, version 11, for Windows (Stata Corporation, College Station, TX).

RESULTS

Cohort and subcohort characteristics.

During the study period, 668 patients in participating hospitals had at least one clinical culture positive for KPCKP. Eleven cases were excluded because of missing data. Four hundred twenty-six (64.8%) of the remaining 657 cases were classified as true infections. Table 1 shows the characteristics of the case cohort and the infected subcohort, each with respective controls. The median (IQR) time at risk for KPCKP isolation in the case cohort was 21 days (10 to 36 days). The 32 (4.9%) cases with index cultures collected within 48 h of admission were all classified as health care associated. Almost half of the index cultures were collected on medical wards (316/657; 48.1%); the other half were fairly equally divided between surgical wards (177/657, 26.9%) and ICUs (168/657, 25.6%).

Analysis of the infected subcohort revealed that the proportion of infections diagnosed on medical wards increased from 37% to 45.4% over the 3-year study period; those diagnosed on surgical wards decreased in frequency (from 31.8% to 22.7%), and the percentage of infections involving ICU patients remained more or less stable (around 30%). In-hospital mortality was significantly higher in the infected subcohort (38.2% [163/426] versus 18.6% [43/231] in the other cases); peak rates were associated with KPCKP bloodstream infections (107/252 [42.5%] versus 56/174 [32.2%] for other infections).

Isolate characteristics.

KPCKP isolates were recovered from blood (252 cases), urine (n = 191), sputum/bronchoalveolar lavage fluid (n = 120), surgical site/wound (n = 62), skin/mucosa (n = 48), cerebrospinal fluid (n = 10), and other cultures (n = 3). In 26 (3.9%) of the 657 cases, a KPC strain was isolated from multiple sites. Most (n = 541 [82.3%]) isolates harbored the bla_KPC-3 gene; the other 116 (17.7%) carried the bla_KPC-2 gene. Approximately half produced extended-spectrum β-lactamases (ESBLs) (CTX-M in most cases). All 657 were resistant to penicillins, cephalosporins, ertapenem, ciprofloxacin, amikacin, cotrimoxazole, and chloramphenicol. Most (65.1%) had meropenem MICs of ≥16 mg/liter, and only 1.4% had MICs of <2 mg/liter for this drug. The vast majority were susceptible to colistin (79.9%), tigecycline (76.4%), and/or gentamicin (81.6%). Seventy-one (10.8%) were susceptible to only one of these drugs (colistin in 47 [7.1%] cases, gentamicin in 12 [1.8%], and tigecycline in 12 [1.8%]). The susceptibility profiles of strains in the infected subcohort were not significantly different from those of strains from other cases.

Risk factors associated with KPCKP isolation or KPCKP infection.

Risk factors for KPCKP isolation were analyzed in 1,971 patients (657 cases and 1,314 matched controls) (Table 1). The eight independent predictors of this outcome identified in logistic regression analysis (i.e., recent admission to ICU, indwelling urinary catheter, central venous catheter [CVC] and/or surgical drain, ≥2 recent hospitalizations, hematological cancer, and recent fluoroquinolone and/or carbapenem therapy) are shown in Table 2. The subgroup analysis (1,278 patients; 426 cases and 852 matched controls) revealed that recent hospitalizations, indwelling CVC, and recent fluoroquinolone and/or carbapenem therapy were also associated with true KPCKP infection, but there were three other factors (i.e., a Charlson index of ≥3, recent surgery, and neutropenia) that specifically predicted KPCKP infection (Table 2). In both analyses, we also replaced the variables related to specific antimicrobial drugs with the variable of recent exposure to ≥3 different antimicrobials. This factor was also independently associated with KPCKP strain isolation (OR, 3.2; 95% CI, 2.35 to 4.30; P < 0.001) and with KPCKP infection (OR, 5.85; 95% CI, 4.04 to 8.50; P < 0.001). In both cases, the risk factors identified as significant when specific antimicrobial use was analyzed remained significant.

TABLE 2.

Logistic regression analysis of risk factors for KPCKP strain isolation and for KPCKP infection

Variablea	OR (95% CI)	P
KPCKP isolation
≥2 previous acute-care hospitalizationsb	5.92 (4.40–7.98)	<0.001
Indwelling central venous catheterc	1.66 (1.29–2.12)	<0.001
Recent carbapenem therapyd	2.98 (2.19–4.05)	<0.001
Recent fluoroquinolone therapyd	1.69 (1.29–2.21)	<0.001
Previous intensive care unit admissionb	5.13 (3.49–7.53)	<0.001
Indwelling urinary catheterc	3.89 (3.03–4.99)	<0.001
Hematological cancer	1.90 (1.27–2.83)	0.002
Surgical drainc	1.62 (1.16–2.45)	0.004
KPCKP infection
≥2 previous acute-care hospitalizationsb	4.26 (3.02–6.01)	<0.001
Indwelling central venous catheterc	2.59 (1.91–3.50)	<0.001
Recent carbapenem therapyd	3.59 (2.46–5.23)	<0.001
Recent fluoroquinolone therapyd	2.22 (1.59–3.10)	<0.001
Charlson score ≥3c	7.49 (5.46–10.27)	<0.001
Recent surgical proceduresd	2.03 (1.48–2.76)	<0.001
Neutropeniac	3.19 (1.50–6.78)	0.003

^aRisk factors identified as significant in this model remained significant (i.e., no significant change in ORs or P values) if the specific antimicrobials were replaced with the variable “recent exposure to ≥3 different antimicrobials.”

^bWithin the 12 months preceding index culture (cases) or at any time during hospitalization (controls).

^cAt the time of the index culture (cases) or at any time during hospitalization (controls).

^dZero to thirty days before index culture (cases) or at any time during hospitalization (controls).

Predictive models for KPCKP strain isolation and for KPCKP infection.

Separate models were developed to predict KPCKP strain isolation and KPCKP infection, based on the data described above. For each model, two versions were constructed, one including the variable of exposure to ≥3 different antimicrobials, the other with the drug class-specific versions. The manner in which recent antimicrobial therapy was analyzed had no significant effect on the AUROCs of either model. However, because both outcomes were strongly associated with different drug use patterns, we report only the models constructed with the drug class-specific variable.

Table 3 shows the distributions of cumulative risk factors among cases and controls in both populations analyzed. Both models displayed good predictive power (Fig. 1). As shown in Table 4, the model for predicting KPCKP isolation performed best with a threshold of ≥3 risk factors (overall accuracy, 79%). The presence of ≥3 risk factors was associated with an OR for KPCKP isolation of 11.33 (95% CI, 8.95 to 14.34; P < 0.001). In the infection prediction model, a threshold of ≥3 risk factors displayed similar predictive accuracy (78%). The presence of ≥3 risk factors was associated with ORs for KPCKP infection of 10.25 (95% CI, 7.57 to 13.91; P < 0.001).

TABLE 3.

Distribution of cumulative risk factors for KPCKP isolation or infection in the respective cohort populations

No. of risk factors	No. (%) of patients
	Isolation			Infection
	Cases	Controls	Total	Cases	Controls	Total
0	45 (8.4)	490 (91.6)	535	9 (2.9)	296 (97.1)	305
1	77 (15.7)	413 (84.3)	490	71 (19.1)	301 (80.9)	372
2	138 (35.1)	255 (64.9)	393	118 (41.1)	169 (58.9)	287
3	146 (57.1)	110 (42.9)	256	119 (63.9)	67 (36.1)	186
4	131 (77.9)	37 (22.1)	168	79 (81.4)	18 (18.6)	97
5	81 (90)	9 (10)	90	26 (96.3)	1 (3.7)	27
6	34 (100)	0 (0)	34	4 (100)	0	4
7	5 (100)	0 (0)	5	0	0	0
8	0	0	0
Total	657 (33.3)	1,314 (66.7)	1,971	426 (33.3)	852 (66.7)	1,278

FIG 1.

Receiver-operating characteristic curves for models predicting KPCKP isolation and KPCKP infection.

TABLE 4.

Performance of the models for predicting KPCKP isolation or infection at different cut-off levels^a

No. of risk factors	TP	FP	TN	FN	Se	Sp	Acc
Isolationb
≥1	612	824	490	45	93	37	56
≥2	535	411	903	122	81	69	73
≥3	397	156	1,158	260	60	88	79
≥4	251	46	1,268	406	38	96	77
≥5	120	9	1,305	537	18	99	72
≥6	39	0	1,314	618	6	100	69
Infectionc
≥1	417	556	296	9	98	35	56
≥2	346	255	597	80	81	70	74
≥3	228	86	766	198	54	90	78
≥4	109	19	833	317	26	98	74
≥5	30	1	851	396	7	100	69

^aTP, number of true positives; FP, number of false positives; FN, number of false negatives; TN, number of true negatives; Se, sensitivity; Sp, specificity; Acc, prediction accuracy rate.

^bTested in 1,971 patients (657 cases with KPCKP isolation and 1,314 matched controls).

^cTested in 1,278 patients (426 cases with KPCKP infection and 852 matched controls).

DISCUSSION

We examined risk factors for the isolation of KPCKP strains in a large population of patients in five large Italian hospitals, where these organisms are an increasing cause of serious infections. To our knowledge, our patient cohort is the largest among published studies specifically investigating the epidemiology and risk factors for KPCKP carriage; in addition, for the first time, we attempted to separately analyze factors associated with KPCKP carriage and infection.

Various types of risk factors were independently associated with KPCKP isolation in our hospitalized patients, including comorbidities (hematological cancer), previous contacts with health care facilities (number of prior hospitalizations and/or previous ICU stay), presence of devices (urinary catheter, central venous catheter, and/or surgical drainage), previous antibiotic treatments, and in particular recent therapy with fluoroquinolones and/or carbapenems. Recent fluoroquinolone therapy was linked to KPCKP isolation in two earlier studies (19, 20); notably, the presence of low-level fluoroquinolone resistance and carbapenem resistance genes (qnrB and bla_KPC-2, respectively) has been demonstrated on the same conjugative plasmid in K. pneumoniae isolates (20, 29). Prolonged treatment with fluoroquinolones could also promote the selection of KPCKP strains by eliminating susceptible competing clones of K. pneumoniae. Similar mechanisms might also explain why recent carbapenem therapy is predictive of KPCKP strain isolation. These drugs have been associated with carbapenem-resistant K. pneumoniae infections and/or isolation of KPCKP in other studies (16–18).

The strongest predictor of KPCKP isolation was a history of ≥2 previous acute-care hospitalizations in the year before the index culture. This variable was not assessed in two of the four earlier studies that specifically investigated KPCKP epidemiology (17, 20). In the third, it showed no association with KPCKP bacteremia (19), and in the fourth, the duration of the previous hospitalization was related to the likelihood of enteric KPCKP colonization at ICU admission (18). ICUs are the principal hospital reservoirs of multidrug-resistant (MDR) bacteria, and a recent history of ICU care markedly increased the probability of KPCKP isolation in our patients (over five times). Papadimitriou-Olivgeris et al. found that it was even more strongly predictive of enteric KPCKP colonization at ICU admission (OR, 12.5) (18).

In our study, KPCKP isolation was also associated with indwelling urinary catheters, CVCs, or surgical drains. The prolonged presence of invasive medical devices like these is a well-known risk factor for colonization or infection by MDR bacteria (5, 30), but it has not been linked to KPCKP isolation in any of the previous studies that looked at these strains (17–20). The same is true for hematological malignancies, whereas cancer in general is a known risk factor for colonization with carbapenemase-producing Gram-negative microbes (5). Some of the risk factors we identified for isolation of KPCKP strains have also been linked to the isolation of ESBL-producing Enterobacteriaceae (30, 31), which is consistent with the high proportion of KPCKP isolates that coexpress ESBLs (roughly half of those in our study).

Accurate risk factor profiles of patients harboring antibiotic-resistant strains (regardless of whether they were the cause of true infections) are useful during the hospitalization assessment for selecting cases that require contact precautions. The model we developed to predict KPCKP isolation provided good discrimination of risk: a threshold of ≥3 risk factors was associated with prediction specificity of 88%. Given its high specificity, the model should be especially effective in identifying high-risk patients. Conventional measures for detecting colonization by resistant bacterial strains (e.g., rectal swabs) and empirical application of appropriate infection control measures could then be limited to this subset of individuals, thereby reducing workloads and costs.

An important feature of our study is the subgroup analysis we carried out to pinpoint specific factors that could be used to identify patients at high risk for true KPCKP infection as opposed to simple colonization. Our retrospective review of each case patient's clinical status at the time of the index culture and the antibiotic regimen that he/she ultimately received indicated that colonization accounted for roughly one third of the cases in which KPCKP was isolated from one or more clinical specimens. The in-hospital mortality in this subgroup was less than half that observed in subsets whose culture positivity reflected true infection (18.6% versus 38.2%, respectively), and this finding indirectly confirmed the accuracy of our classification. Analysis of the cases classified as true KPCKP infection revealed seven variables that were significantly associated with this outcome (a Charlson index of ≥3, indwelling central venous catheter, recent surgical procedures, neutropenia, ≥2 recent hospitalizations, and recent fluoroquinolone and/or carbapenem therapy).

These factors were used to develop a model for predicting KPCKP infection, which also displayed good discriminatory power. Its use might improve decisions regarding empirical antimicrobial therapy, particularly in settings where these infections are endemic. However, it is important to note that four of the seven variables included in this model were also predictive of the broader outcome, i.e., KPCKP strain isolation. This is not particularly surprising, because the infected patients represented about two thirds of all the patients with KPCKP culture positivity. It means, nonetheless, that in a certain proportion of patients, the model will not be able to discriminate between isolation and infection by KPCKP. With this limitation in mind, however, clinicians working in hospitals where KPCKP is epidemic/endemic might still find this second model useful when empirical antimicrobial therapy is being prescribed for a patient with clinical signs of severe infections. Indeed, it could potentially reduce the risk of ineffective therapy during the initial phase of treatment, a major risk factor for mortality in patients with severe infections in general (30, 32–35) and in particular those suffering from bacteremia caused by KPCKP (7). As always, of course, empirical therapy must be reviewed as soon as in vitro susceptibility data are available and promptly de-escalated when appropriate.

One of the limits of our study is that it did not eliminate the potentially confounding effects of patients' proximity to others harboring KPCKP (i.e., those in adjacent beds). However, it seems unlikely that the importance of this factor in determining the risk for KPCKP isolation or true infection was seriously underestimated, since cases and controls were matched for both ward and month of admission. It is also important to recall that rectal swab screening for KPCKP colonization was not routinely performed on admission in any of the hospitals that took part in this study. Therefore, we cannot exclude the possibility that one or more control patients had asymptomatic KPCKP colonizations. If so, this might have led to an underestimation of the importance of certain risk factors.

In conclusion, our results can be useful for identifying patients at high risk for harboring KPCKP isolates and for containing the spread of these organisms in hospitals. Additional studies are needed to define which KPCKP-colonized patients are most likely to develop true infections during their hospitalization.