The Likelihood of Hospital Readmission among Patients with Hospital-Onset Central Line-Associated Bloodstream Infections

Abstract

Objective

We sought to determine whether central line-associated bloodstream infections (CLABSI) increase the likelihood of readmission.

Design

Retrospective matched cohort study for the years 2008–2009.

Setting

Acute care hospitals.

Participants

Medicare recipients. CLABSI and readmission status were determined by linking National Healthcare Safety Network surveillance data to the Centers for Medicare & Medicaid Services’ Medical Provider and Analysis Review in eight states. Frequency matching was used on ICD-9-CM procedure code category and intensive care unit status.

Methods

We compared the rate of readmission among patients with and without CLABSI during an index hospitalization. Cox proportional hazard analysis was used to assess rate of readmission (the first hospitalization within 30 days post-index discharge). Multivariate models included the following covariates: race, sex, length of index hospitalization stay central line procedure code, GAGNE co-morbidity score, and individual chronic conditions.

Results

Of the 8,097 patients, 2,260 were readmitted within 30 days (27.9%). The rate of first readmission was 7.1 events/person-year (PY) for CLABSI patients and 4.3 events/PY for non-CLABSI patients (p <0.001). The final model revealed a small but significant increase in the rate of 30 day readmissions for patients with a CLABSI compared to similar non-CLABSI patients. In the first readmission for CLABSI patients, we also observed an increase in diagnostic categories consistent with CLABSI including septicemia and complications of a device.

Conclusions

Our analysis found a statistically significant association between CLABSI status and readmission, suggesting that CLABSI may have adverse health impact that extends beyond hospital discharge.

Introduction

Readmissions to acute care hospitals create a burden for patients and their health, accounting for increased costs, resources, and time for healthcare providers, payers, and ultimately the healthcare system.

Despite making progress, healthcare-associated infections (HAI) continue to impact patients in the United States. One in 25 hospital patients develop at least one HAI during hospitalization.¹ Estimates suggest HAIs result in $28 to $34 billion in excess healthcare costs each year.² Although the number of central line-associated bloodstream infections (CLABSI) has decreased over the last decade,^3,4 it is estimated over 30,000 occur nationally in hospital wards and critical care units. CLABSIs may lead to longer hospital stays, increased mortality, and increased costs.^5–8

Readmissions or re-hospitalizations are challenging because they occur frequently and are costly to payers such as Medicare.^9,10 Rates of hospital readmission among adults can vary from five to 29%^11–15 and are responsible for up to 60% of hospital expenditures.¹⁶ Prior research indicates there exists an association between having an HAI and becoming re-hospitalized. In one single center study, HAI incidents were the cause of 14.3% of readmissions.¹⁷

The ongoing problem of hospital readmissions continues to result in serious public health consequences by creating a burden on patients and generating unnecessary healthcare costs. Previous studies of CLABSI have focused on the visit in which the CLABSI occurred and do not examine the issue of readmission. The purpose of this analysis was to determine whether an association exists between patients identified as having a CLABSI and subsequent readmission to acute care hospitals.

Methods

We conducted a retrospective cohort study for the years 2008–2009 to compare the rate of hospital readmissions among those with a hospital-onset CLABSI to frequency matched controls. Since, as previously shown, ICD-9-CM codes are not sufficiently able to identify CLABSI cases in administrative files,¹⁸ we linked data from the National Healthcare Safety Network (NHSN) to identify CLABSI cases among a population of hospitalized Medicare enrollees identified from the Medicare Provider Analysis and Review (MedPAR) database and Beneficiary Annual Summary File (BASF) obtained from the Centers for Medicare & Medicaid Services (CMS). The methods for linking NHSN and MedPAR datasets to identify individual Medicare recipients with an HAI and frequency matching of uninfected patients have been previously described.^19–21 The CDC Human Research Protection Office determined this work was exempt from the regulations under 45 CFR 46.101(b)(5). This work was conducted under a data use agreement with CMS.

Data Sources and Linkage

Cases of hospital-onset CLABSI were identified using data extracted from the NHSN CLABSI surveillance module for admissions. CLABSIs were defined according to the standard NHSN protocol. Laboratory-confirmed bloodstream infections not secondary to another HAI were considered to be central line associated if a central line or umbilical catheter was in place at the time or within 48 hours before the onset of the infection.²² Admission dates, date of birth, sex, facility and its location, and date of infection were captured electronically. For this analysis, only MedPAR data from 2008–2009 were available to us from Colorado, Illinois, New Hampshire, New York, Pennsylvania, South Carolina, Tennessee, and Virginia, chosen due to their participation in NHSN. The MedPAR database contains claims for beneficiaries from certified inpatient hospitals and skilled nursing facilities.²³ Only claims from inpatient hospital stays were included. This database along with the BASF provided information on admission dates, date of birth, sex, and facility for linking as well as information on diagnosis and procedures using ICD-9-CM codes, reimbursement cost of the claim, beneficiary status, and CMS chronic conditions. An encrypted beneficiary identifier was available in to order to follow beneficiaries longitudinally and determine their readmission status including readmission to other facilities. Unique healthcare facilities from the NHSN facility file were mapped to the CMS provider ID using the reported CMS provider ID when available or facility name and location from NHSN and data from the CMS Cost Reports, 2004–2009.

For both data sources, we limited the population to those over the age of 64 with a valid date of admission from January 2008 through December 2009, a valid date of birth, sex, and facility. In the MedPAR file, patients were also limited to those who aged into the Medicare recipient cohort with or without end stage renal disease, were enrolled in Medicare Part A and B throughout their eligibility, and never enrolled in a Medicare Advantage (HMO) program. We also eliminated hospital visits to certain special units such as psychiatric and swing units. To identify individuals with CLABSI, CLABSI events reported to NHSN were linked to hospital claims data in MedPAR using a combination of four variables including hospital admission date, date of birth, sex, and unique facility identifier.^19,21 Only unique, exact matches among those variables were included in the analysis.

Control Selection and Frequency Matching

To control for potential counfounding, first, potential controls were limited to the same facilities, age, primary ICD-9-CM diagnoses, and diagnosis-related groups (DRGs) observed in the population of patients with CLABSI. Patients with a diagnosis consistent with CLABSI (ICD-9-CM 999.31) but not identified as NHSN cases were eliminated from the potential control pool. Second, five non-CLABSI control stays were selected such that the frequency of the primary ICD-9-CM procedure category, based on single level Clinical Classification Software (CCS) available from the Agency for Healthcare Research and Quality (http://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp),²⁴ and intensive care unit (ICU) status were similar between CLABSI stays and non-CLABSI stays.²¹ Therefore, patients with CLABSI (as reported to NHSN) made up our exposed group, while frequency matched controls without CLABSI made up our unexposed group for comparison.

Outcome

Hospital readmissions ocurring 1–30 days after the initial hospital discharge (index hospitalization) represented the primary outcome of the study. Patients discharged from their index hospitalization and readmitted on the same day were considered transfers and excluded from the analysis along with patients who died during the index hospitalization. We also considered the first hospital readmission after the index hospitalization regardless of timing of readmission as a secondary outcome.

Statistical Analysis

For univariate analysis, we used χ² test for dichotomous outcome measures, and Fisher’s Exact test as appropriate. T-tests and Wilcoxon-rank sum were used for continuous variables. Potential confounders and interaction terms were assessed in both stratified and multivariable analyses. Potential interaction terms were assessed in the stratified analysis using the Breslow-Day test.

To assess the association between CLABSI and the rate of first re-hospitalization, the rate of initial readmission among those with a CLABSI and those without was compared through survival analysis using a Cox proportional hazard model. Patients were censored at death or the end of 30 days for the primary analysis. For our secondary outcome, time was allowed to accumulate from the time of discharge until readmission, death, or the end of the study period. Multivariate models included terms for age, race, sex, index hospitalization length of stay (LOS), presence of an ICD-9-CM procedure code for insertion of a central line, individual CMS chronic conditions, and indication of ICU care. These terms were based on the patient’s index hospital stay. Given that the co-morbidity score²⁵ violated the proportional hazards assumptions, models were stratified by co-morbidity score. The final model was determined by assessing potential confounding in the multivariate models using methods previously described by Kleinbaum, et. al.²⁶ For the final models, all the terms but the CMS chronic conditions terms were included. In addition, because certain terms were found to be significant effect modifiers by the Breslow-Day test in the stratified analyses, a secondary analysis was conducted to evaluate those terms in multivariate Cox proportional hazard models. Two significant interaction terms, index hospitalization LOS and the CMS chronic condition rheumatoid arthritis, were included in the final model. Confounding was assessed as previously, and all terms including the CMS chronic conditions were included in those models.

In addition, to compare the potential reasons for readmission, we examined the frequency of the most common ICD-9-CM primary diagnosis category for the first readmission visit by CLABSI status. Differences in frequencies by CLABSI status were assessed using χ² test or Fisher’s Exact test as appropriate. We also examined the frequency of patients discharged and readmitted within the same day by CLABSI status along with admission type for the subsequent stay.

Analyses were conducted using SAS 9.3 statistical software (SAS Institute, Cary, NC). Alpha was set to 0.05 for all statistical analyses.

Results

MedPAR and NHSN data from eight states were linked to determine which individuals in the MedPAR dataset experienced a CLABSI during hospitalization. In those eight states, there were over 3.95 million MedPAR records available and 4,747 CLABSI events among patients >64 years of age reported to NHSN. Of those 4,747, 41% (1,967) of NHSN records linked to a MedPAR inpatient hospital claim record. Given the proportion of persons >64 who use Medicare as the primary payer, who are enrolled in both Parts A and B and not the Medicare Advantage program and have aged into the Medicare cohort, only 49.8% or 2,364 NHSN events were expected to link. Therefore, our adjusted linkage rate is 83.2% (1,967/ 2,364). After limiting potential controls to those claims with same facilities, age range, range of primary ICD-9-CM diagnoses, and range of DRGs, 1.05 million non-CLABSI patients remained eligible to be selected as controls. After frequency matching, 9,835 controls were randomly selected for 1,967 cases, resulting in a total of 11,802 patients selected for the study. For all selected patients, hospital readmissions after discharge from the index hospitalization were identified.

Among the 11,802 patients, 8,097 patients survived the index hospitalization and were not transferred or re-hospitalized on the same day of discharge (Table 1). Among the 8,097, 917 (11.3%) had a CLABSI during the index visit and 7,180 did not have a CLABSI (88.7%). Demographics and clinical characteristics varied among those with and without a CLABSI (Table 1).

Table 1.

Demographic and Clinical Characteristics of Study Population

	Non-CLABSI		CLABSI
	n=7180	88.7%	n=917	11.3%	p-value†
Follow up time††	447.7		48.0		----
Readmitted within 30 days	1920	26.7%	340	37.1%	<0.0001
Race					0.0001
White	6001	83.6%	739	80.6%
Black	934	13.0%	160	17.5%
Other	245	3.4%	18	2.0%
Sex					0.654
Male	3749	52.2%	486	53.0%
Age§	77.1	7.7	75.5	7.4	<0.0001
ICU Status§§	5035	70.1%	626	68.3%	0.248
Central line procedure code§§	1850	25.8%	472	51.5%	<0.0001
Length of stay§§					<0.0001
0–6 days	2157	30.0%	13	1.4%
7–12 days	2461	34.3%	71	7.7%
13–22 days	1618	22.5%	241	26.3%
over 22 days	944	13.2%	592	64.6%
Died within 30 days of discharge	818	11.4%	192	20.9%	<0.0001
Died after 30 days of discharge	1632	22.7%	241	26.3%	0.016
Co-morbidity score§§					<0.0001
NA	420	5.9%	129	14.1%
−1	369	5.1%	19	2.1%
0	586	8.2%	21	2.3%
1	1327	18.5%	154	16.8%
2	1224	17.1%	212	23.1%
3	1147	16.0%	163	17.8%
>4	2107	29.4%	219	23.9%
Alzheimer's disease‡	632	8.8%	72	7.9%	0.336
Alzheimer's disease and related disorders‡	1563	21.8%	204	22.3%	0.742
Acute myocardial infarction‡	289	4.0%	40	4.4%	0.627
Atrial fibrillation‡	1588	22.1%	218	23.8%	0.257
Breast cancer‡	149	2.1%	23	2.5%	0.392
Cataracts‡	1281	17.8%	149	16.3%	0.234
Chronic heart failure‡	3712	51.7%	525	57.3%	0.002
Chronic obstructive pulmonary disease‡	2311	32.2%	307	33.5%	0.431
Depression‡	1358	18.9%	202	22.0%	0.024
Diabetes‡	3157	44.0%	435	47.4%	0.047
Endometrial cancer‡	25	0.4%	≤10¥		0.354
Glaucoma‡	647	9.0%	75	8.2%	0.405
Hip/pelvic fracture‡	107	1.5%	14	1.5%	0.932
Ischemic heart disease‡	4825	67.2%	620	67.6%	0.803
Chronic kidney disease‡	2842	39.6%	457	49.8%	<0.0001
Lung cancer‡	237	3.3%	31	3.4%	0.899
Osteoporosis‡	891	12.4%	90	9.8%	0.023
Prostate cancer‡	365	5.1%	39	4.3%	0.277
Rheumatoid Arthritis‡	1832	25.5%	242	26.4%	0.568
Rectal cancer‡	346	4.8%	45	4.9%	0.906
Stroke or transient Ischemic attack‡	840	11.7%	128	14.0%	0.047

^†P-values based on Chi-square tests

^††Values expressed as person-years (PY)

^§Values expressed as mean (standard deviation)

^§§During the index CLABSI visit

^‡CMS Chronic Conditions Categories

^¥In accordance with the CMS data use agreement, the actual number and percentage were not displayed for cell sizes ≤10.

Overall, 2,260 of these patients (27.9%) were readmitted within 30 days (Table 1). Of the 917 with CLABSI, 340 (37.1%) were readmitted within 30 days compared to 26.7% among non-CLABSI, p<0.0001. The rate of readmission within 30 days was 7.1 events per person-year (PY) for those with CLABSI. Among non-CLABSI patients, the rate of readmission in 30 days was 4.3 events/PY. Therefore, the rate of readmission within 30 days was 1.7 times higher among CLABSI patients compared topatients without a CLABSI [IDR=1.7, 95%, CI (1.5, 1.9)]. In addition, 550 (60.0%) of those with a CLABSI and 3,962 (55.2%) of those without a CLABSI were ever readmitted during the study period, resulting in overall readmission rates of 2.5 events/PY and 1.4 events/PY respectively. The overall rate of readmission was 1.8 times higher for those with a CLABSI [IDR=1.8, 95% CI (1.6, 1.9)].

When adjusting for potential confounders, our Cox proportional hazards model demonstrated a borderline significant association between CLABSI and 30 day readmission [IDR=1.2, 95% CI (1.0, 1.3)]. We also observed a borderline association between CLABSI and all readmissions in the Cox proportional hazards model [IDR=1.1, 95% CI (1.0, 1.2)]. In addition, we examined the ten most common primary ICD-9-CM discharge CCS categories for the first readmission visit by CLABSI status (Table 2). Septicemia was the most common diagnosis category for readmission among those with a CLABSI and was reported over twice as often compared to those without CLABSI, p<0.0001. Complications of a device, urinary tract infections, and intestinal obstruction without hernia were also more commonly reported among those with a CLABSI. In total, the most common primary categories account for 63% of the readmission events among those with a CLABSI.

Table 2.

Most Frequent Primary ICD-9-CM Discharge CCS Category^* for the First Readmission Visit by CLABSI Status

CCS Category	Non-CLABSI		CLABSI
Septicemia†	241	3.4%	63	6.9%
Complications of device †	83	1.2%	31	3.4%
Complications of surgical procedure	174	2.4%	26	2.8%
Pneumonia	97	1.4%	17	1.9%
Congestive heart failure	131	1.8%	13	1.4%
Respiratory failure; insufficiency	94	1.3%	12	1.3%
Urinary tract infections ‡	41	0.6%	11	1.2%
Aspiration pneumonia	59	0.8%	≤10¥
Acute renal failure	63	0.9%	≤10¥
Gastrointestinal hemorrhage	53	0.7%	≤10¥
Intestinal Infection	42	0.6%	≤10¥
Intestinal obstruction without hernia ‡	21	0.3%	≤10¥

^†p < 0.0001;

^‡p < 0.05

^¥In accordance with the CMS data use agreement, the actual number and percentage were not displayed for cell sizes ≤10.

Note: Table includes 12 CCS categories since for CLABSI, the 10^th through 12^th categories each contained an equal number of events.

^*- http://www.hcup-us.ahrq.gov/toolssoftware/ccs/ccs.jsp

In our secondary analysis which included potential effect modifiers, the final stratified Cox models demonstrated a statistically significant association between CLABSI and readmission to an acute care hospital modified by the effect modifiers LOS and rheumatoid arthritis (Table 3). As the index visit’s LOS decreased, the rate of readmission for those with a CLABSI increased compared to those without a CLABSI. For example, for CLABSI patients with LOS ≤ 6 days, the rate of readmission within 30 days was 5.5 to 7.5 times greater than patients without a CLABSI. A positive history of rheumatoid arthritis also increased the hazard ratio for readmission. However, for patients without a history of rheumatoid arthritis and a LOS > 6 days, the rate of readmission within 30 days was not significantly higher for CLABSI patients compared non-CLABSI patients. When analyzing all readmissions during the study period, the hazard ratios decreased slightly compared to 30 day readmissions.

Table 3.

Final hazard ratios and 95% CI for CLABSI at the specific levels for the interaction terms, length of stay and rheumatoid arthritis, among patients who were readmitted within 30 days of the index hospitalization and for time to first readmission after index hospitalization

Readmission within 30 Days of the Index Hospitalization
	Rheumatoid arthritis
	Yes		No
Length of stay§§	IRR	95% CI	IRR	95% CI
0–6 days	7.5	(4.0 14.4)	5.5	(2.9 10.6)
7–12 days	1.9	(1.2 3.0)	1.4	(0.9 2.2)
13–22 days	1.4	(1.0 1.9)	1.0	(0.8 1.3)
over 22 days†	1.3	(1.0 1.7)	1.0	(0.8 1.2)
First Readmission Regardless of Timing after Index Hospitalization
	Rheumatoid arthritis
	Yes		No
Length of stay§§	IRR	95% CI	IRR	95% CI
0–6 days	3.8	(2.1 7.0)	3.0	(1.6 5.5)
7–12 days	1.6	(1.1 2.3)	1.3	(0.9 1.8)
13–22 days	1.3	(1.0 1.6)	1.0	(0.8 1.2)
over 22 days†	1.3	(1.0 1.5)	1.0	(0.9 1.1)

^§§During the index CLABSI visit

^†Reference category

^*Cox proportional hazards models were stratified by Gagne co-morbidity score and included age, sex, race, ICU status, central line procedure code, and CMS chronic conditions as co-variates. Terms for length of stay (LOS) and the CMS chronic condition rheumatoid arthritis and the effect modification by CLABSI for both of those terms were also included in the model.

While our primary analysis eliminated those patients who were discharged and readmitted on the same day, patients with a CLABSI were more likely to be discharged and readmitted on the same day. Of the 1,239 CLABSI cases that did not die in the hospital, 321 (25.9%) were discharged and readmitted on the same day while only 856 (10.6%) of the 8,055 non-CLABSI cases were readmitted, p<0.0001. Of those admitted on the same day as discharge, those with a CLABSI were more likely to have an urgent or emergency readmission (33% vs. 23%, p=0.0004).

Discussion

In our study, CLABSI was determined to be significantly associated with readmission to an acute care hospital. Further, readmission rates were highest among patients with shorter LOS during their index visit or a history of rheumatoid arthritis. While our study focuses specifically on CLABSI, our findings are consistent with previous studies suggesting that HAIs may increase the risk of re-hospitalizations and have adverse health impact and burden that extends beyond hospital discharge.^17,27–29

It is important to note there are few studies specifically examining the issue of CLABSI and hospital readmissions. Although the results are statistically significant, further research is needed, especially among different populations to determine the consistency of these findings. If confirmed, our findings further reinforce the need to prevent CLABSI, as this may benefit beyond the index visit though the total burden of readmissions attributable to CLABSI may not be large.

Our analysis also identified LOS and a diagnosis of rheumatoid arthritis as potential effect modifiers. Few studies have established relationships between re-hospitalization and LOS, as well as between re-hospitalization and rheumatoid arthritis. Kaboli et al. concluded that patients with an increased LOS had a higher likelihood of readmission, a three percent increase for every one extra day of stay.³⁰ In our analysis, a longer LOS during the index hospital stay was also associated with a higher rate of re-hospitalization (data not shown), but if a patient was exposed to CLABSI in the index hospitalization, a shorter LOS indicated a higher rate of readmission compared to those without a CLABSI but similar LOS (Table 3). We also found that exposure to CLABSI increased rate of re-hospitalization among patients diagnosed with rheumatoid arthritis (Table 3). The reasons for this association are unclear. Future studies could examine the potential interaction between LOS and rheumatoid arthritis with readmission following CLABSI as well as possible mechanisms for effect modification.

Our study had a number of strengths. First, NHSN and MedPAR data were linked to identify patients who were both infected with CLABSI and re-hospitalized. Therefore, we did not depend on the patients’ ICD-9-CM codes to identify CLABSI, which previous research has shown their inability to properly differentiate HAIs.^30,31 In fact, one study found that administrative data often misclassified non-CLABSI cases as true CLABSI cases, producing a different number of cases compared to that of surveillance data.¹⁸ Studies using only ICD-9-CM codes for identification of CLABSI suffer from strong misclassification bias in determining the exposure status. Also, data for readmissions were based on beneficiary claims in the MedPAR dataset, which are reliable for identifying longitudinal visits for beneficiaries even across different facilities, and also provide additional demographic and clinical information valuable for risk adjustment. Further, when we examined the primary diagnosis code of the first readmission, among those with a previous CLABSI, we observed an increase in diagnostic categories consistent with CLABSI including septicemia and complications of a device.

A limitation of our analysis is the inability to differentiate between a true, unplanned re-hospitalization and a planned hospital visit following discharge. It is possible or even likely that some of the readmissions included in the analysis represent planned readmission. However, the frequency of planned readmission would not be expected to have varied by CLABSI status given the frequency matching in the cohort. Hence any potential bias would be non-differential and bias towards the null. Further, because administrative data was used, we are unable to specifically determine how the preceding CLABSI was potentially related to the increase in the rate of readmissions. In addition, administrative data are not collected for research purposes, and therefore, misclassification may occur for other data derived from the MedPAR data source.^32,33 Another limitation is the potential for mismatches in the NHSN and MedPAR data linkage. By using specific requirements and allowing for only exact matches among our linkage variables, the likelihood of a mismatch is rare. Further since CLABSIs are rare and we eliminated patients who did not link but had an ICD-9-CM code consistent with CLABSI, it is unlikely our controls experienced a CLABSI. Again, such misclassification would have biased our results toward the null. Additionally, while we attempted to control for confounding through matching and multivariate models, there is potential for unmeasured confounding to exist in our analysis given the availability of data elements in our data sources. Finally, as the finding of the effect modifiers LOS and rheumatoid arthritis were unexpected, our control selection did not take this finding into account, and while our results would not be expected to be biased, future studies of readmission and HAI should consider the role of LOS into the design of the study.

In conclusion, our study found a significant association between CLABSI and the risk of readmission to an acute care hospital. Prevention of CLABSI may therefore reduce patient burden and healthcare costs associated not only with hospitalizations during which CLABSIs occur,^2,33,34 but also by prevention of a proportion of subsequent readmissions to the hospital and their associated costs.