Risk Factors for Urinary Tract Infection (UTI) Among Home Health Care Patients: Analysis Using Routinely Collected Clinical Data

Background: Urinary tract infection (UTI) is a complication often experienced during a home health care (HHC) episode, yet related hospitalization risk factors are unclear. Objectives: This study use multiple data sources to identify risk factors for UTI related hospitalization or emergency department (ED) visits among HHC patients. Method: We performed a multivariable logistic regression to identify risk factors for UTI-related hospitalization or ED visits using merged data from the Outcome and Assessment Information Set, electronic health record from a large HHC agency, and Center for Medicare and Medicaid claims. Results: Of 48,336 cases, 1,689 patients (3.5%) had a UTI-related hospitalization/ED visits. Being a female (OR = 1.31; 95% CI: 1.16–1.46), the presence of a urinary catheter (OR = 5.7; 95% CI: 4.54–7.14), treatment with general antibacterial and antiseptics (OR = 2.75; 95% CI: 1.02–7.38), dependency in instrumental activities of daily living (e.g., meal preparation [OR=1.72; 95% CI: 1.25-2.37]), and no available caregivers (OR = 1.79; 95% CI: 1.2–2.68) increased the odds of a UTI related event among HHC patients. Discussion/conclusion: We identied notable risk factors for UTI related hospitalization/ER visit, lling a knowledge gap on the currently understudied HHC population. Risk factors identied in this study can be used to proactively identify HHC patients at risk for UTI related hospitalization and target them for preventive interventions. Further research is needed in HHC to develop tailored interventions for at-risk patients.


Background
Infections are a substantial burden in home health care (HHC), but there is limited research on risk factors that could inform early detection and prevention measures. A study using a national data sample reported that 18.2% of unplanned hospitalizations among HHC patients were caused by four types of infections (e.g., respiratory, wound, urinary tract, and intravenous (IV) catheter-related), making infections as one of the top reasons for unplanned hospitalization among HHC patients [1]. Another study reported that approximately 1,133 of 24,887 U.S. HHC patients or 4.6% had UTI-related hospitalizations [2]. In general, it is well established that UTIs are one of the most common healthcare-associated infections, accounting for up to 36% of all healthcare-associated infections [3]. One study reported that 8.5% of patients developed catheter-associated urinary tract infections in community settings [4].
Early detection and treatment of UTI is critical as delayed intervention can result in serious and lifethreatening complications, such as renal damage or sepsis [5]. One recent study examining UTI related hospitalizations/ emergency department (ED) visits in HHC, reported that some of the main UTI related hospitalization risk factors included female sex, being a Medicaid recipient, severe di culties in performing activities of daily living (ADL), the presence of a primary caregiver to provide assistance with various functional tasks, and the use of a urinary catheter [6]. However, this study only used Outcome and Assessment Information Set (OASIS) data, which is based on clinicians' report in measuring infection outcomes, so may underreport UTIs. Besides, it only included a limited set of pre-selected variables and excluded other potential risk predicting variables, such as the patient's diagnoses and vital signs.
To address the knowledge gaps and advance the rigor of previous research, this study is aimed to identify factors predicting HHC patients' risk of UTI-related hospitalization or ED visits using a combination of data sources. We complemented OASIS data with routinely collected data from agency's electronic health records (EHR) and outcome data extracted from the Center for Medicare and Medicaid (CMS) claims.
EHR from agency collects rich clinical risk factors such as patients' vital signs and medication regimens. Combining these important previously underutilized data sources allows us to identify a greater array of risk factors, which can lead to more re ned risk modeling.

Methods
This was a retrospective cohort study. This study was approved by the Institutional Review Boards of our institution and the agency from which the patient data were obtained.

Sample and datasets
The study sample included all adult patients (managed care patients were excluded) served by a large HHC agency from January 1 to December 31, 2014. Data included the OASIS assessment at the start and end of HHC, as well as additional clinical and administrative data documented in the agency's EHR. We obtained Medicare claims data from CMS for all patients in the sample who had standard Medicare fee for service coverage. Data were merged at the patient and episode levels using patients' study ID and case sequence number.

Outcome and Assessment Information Set (OASIS)
The OASIS is the only standardized HHC patient assessment mandated by the CMS for all Medicarecerti ed HHC agencies nationwide. Each HHC patient is assessed by a HHC clinician using OASIS 1) at the time of admission and readmission-called start of care and resumption of care in OASIS, respectively; 2) at the time of discharge from HHC-called end of care, or during transition from HHC to other levels of care-e.g., hospital within an HHC episode as needed-designated as a transfer. OASIS tracks nearly 100 HHC patient characteristics in the domains of socio-demographics, medical history, health status, environmental status, support system, functional status, and health service utilization. Researchers have found moderate to excellent reliability for certain OASIS items [7] such as ADLs, Instrumental Activities of Daily Living (IADLs), clinical items, and behavioral assessment [7,8].

CMS claims data
The CMS claims data were used to identify UTI related hospitalizations or ED visits. There are four main types of Medicare claims data: bene ciary enrollment information, Part A, Part B, and Part D. The outcome data used in this study are from Part A service utilization, which includes information from inpatient utilization such as summary information from hospitalizations or ED visits and detailed hospital claims.

Outcome variables
The outcome variable of this study is hospitalizations or ED visits due to UTI occurring up to 60 days after HHC admission as identi ed by CMS claims data. Table 1 shows the seven International Classi cation of Diseases (ICD 9) codes that were used to identify a UTI-related hospitalization or ED visits for this study. OASIS variables from start of care and data from the agency's patient administrative and EHR dataset were used as independent variables. OASIS variables were operationalized as needed. For example, items that allowed multiple selections ("Mark all that apply") were operationalized into multiple binary variables. Vital signs at admission together with medication therapeutic classes were extracted and merged into the study dataset. The disease classi cations were speci cally developed for HHC patients [9]. Body temperature was categorized as high ( > = 100.4°F) or normal (< 100.4°F). For more examples of how the independent variables were operationalized for this study, see Table S (Table S., Additional le 2).

Data analysis
A nal dataset was constructed by merging data sources using the study ID crosswalk that was established for each data source (the unique identi er) along with a case sequence number (some patients had multiple HHC admissions during the study timeframe). The sample was randomly divided into two sub-samples with 70% of the data used for model development (training data) and the other 30% for model testing (test data) [10]. Given that the total sample is over 48K, we had enough power to detect the statistically signi cant differences between the groups.
In our analysis, we rst examined distributions of study variables in the training data. For categorical variables with counts of the event in a cell of less than 20 observations, we collapsed the cells by combining nearby categories for more accurate estimation. Secondly, we conducted bivariate analysis and selected signi cant variables associated with UTI events using a criterion of p < 0.2 for entry into an initial "maximal" multivariable logistic regression model. Then, we performed multivariable logistic regression to build the model using stepwise variable selection technique, which helped us to identify the most predictive variables. Finally, we t the model to the test data and evaluate the model's predictive performance using the area underneath the receiver operating characteristic curve (AUC).

Results
We performed a descriptive analysis of the study patient sample (N = 48,336 cases) and compared the patients with and without a UTI-related hospitalization or ED visits (

Discussion
The purpose of this study was to explore risk factors associated with UTI related hospitalization or ED visits during HHC. The strength of this study is its use of claims data to objectively identify UTI related hospitalization or ED visits. This study identi ed several signi cant risk factors in the domains of previous medical history, diagnoses, risk for hospitalization, elimination status, ADL/IADLs, medication, and vital signs.
Some of our ndings con rm what has been reported in previous studies. Being a female, presence of a urinary catheter, and dependency in ADL/IADLs increased the odds of UTI related hospitalization or ED visits among HHC patients [6]. History of UTI treatment within 14 days before admission has also been previously reported as associated with a UTI-related hospitalization or ED visits [6] and is con rmed by our study. High temperature was reported to be a signi cant factor for UTI related ED return visits in another study [12], and was found in the present study to be associated with a UTI-related hospitalization or ED visits.
Several medications were associated with UTI related hospitalization or ED visits including antibiotic regimens. General antibacterial and antiseptics were associated with increased odds of UTI-related hospitalization or ED visits. This is an anticipated result, since patients taking these medications could have UTIs, thus are at higher risk for UTI-related hospitalization or ED visits. However, the con dence interval is wide from 1.02 to 7.38 due to small number of patients who were treated with these medicines, we should be cautious on interpreting the results.
Having a problematic observable surgical wound was found to be associated with decreased odds of UTI-related hospitalization or ED visits. Patients with problematic observable surgical wound had 37% lower odds of UTI-related hospitalization or ED visits, compared to patients without such wounds. Patients with wounds are likely to have frequent visits from HHC nurses to manage wounds [13] and might be under closer observation for general infection signs and symptoms, which may facilitate early detection of UTI and therefore prevent related hospitalizations or ED visits. In general, higher disease complexity yields closer clinical observation; therefore, a higher chance of detecting any early signs of infections.
Although rigorous research and practice has focused on preventing UTIs for patients using a urinary catheter in various settings [14][15][16][17], having a urinary catheter was the most signi cantly associated factor for UTI related hospitalization and ED visits prediction. Notwithstanding the efforts of nurses and administrators to prevent UTI-related hospitalization or ED visits among patients with an indwelling urinary catheter (e.g., through reminding the nurses, aseptic insertion) [4,16,17], UTI-related hospitalization or ED visits remains still high in this speci c group. More rigorous research and improvement in clinical practice in this area are much needed.
Under the care management category, likelihood of assistance from informal caregivers was associated with odds of UTI related hospitalization and ED visits. Patients who reported no available caregivers outside of the HHC agency to assist them were associated with increased odds of UTI events. This nding is concordance with previous studies that reported family caregiver involvement reduced catheterassociated urinary tract infection [18]. Informal, non-agency, caregivers are the majority of home-based patient caregivers around the world, but their potential contributions to mitigating infection risk at home has not been considered. Informal caregivers are often indispensable for providing care to patients at home [19] and the Association for Professionals in Infection Control and Epidemiology have emphasized the critical impact of caregivers on HHC infection control [20]. Thus, informal caregiver availability and likelihood to provide care to homecare patients has to be considered in care management of patients in order to reduce UTI related hospitalization and ED visits. Finally, our model showed good predictive performance in identifying HHC patients at risk for a UTI event.
In settings other that HHC, mostly hospitals, such models are being currently used as early warning systems to inform identi cation of patients at risk [21,22]. Our results show that it is feasible to create a UTI risk model in HHC but further research is needed to implement and test this model in clinical practice.

Limitations
This study has several limitations. First, the study was conducted using data from only one HHC agency, which mostly serves an urban population with not-for-pro t ownership, and we excluded managed care patients due to merging with claims data so the ndings might be not widely generalizable. In addition, we only used a single year of data, although it includes a large number of episodes over a full year. There remains a possibility that the observations might have been affected by circumstances particular to the year studied.

Implications
To develop UTI early detection and prevention practice in HHC, it is important to identify HHC patients who are at high risk for UTI related hospitalization and ED visits. Female patients with a fever, presence of a urinary catheter, history of UTI at HHC admission, and certain antibiotic medications should be agged for special closer observation and follow-up. Building on our ndings, we hope future researchers in HHC can develop more precise analytical models and utilize them to successfully identify HHC patients at high risk for UTI related hospitalization.

Conclusion
This study adds to the limited body of knowledge on risk factors associated with UTI related hospitalization or ED visits among patients receiving HHC services. Using merged data from OASIS, EHR from one large HHC agency, and CMS claims, we found several risk factors. Better knowledge of risk factors and attention to predictive factors, e.g. medication regimen at HHC admission, can inform better prediction models of patients' hospitalization or ED visits related to UTI, which can lead to improved case management and reduced care costs.