Personalized infection prevention and control: identifying patients at risk of healthcare-associated infection.

BACKGROUND
Few healthcare-associated infection (HAI) studies focus on risk of HAI at the point of admission. Understanding this will enable planning and management of care with infection prevention at the heart of the patient journey from the point of admission.


AIM
To determine intrinsic characteristics of patients at hospital admission and extrinsic events, during the two years preceding admission, that increase risk of developing HAI.


METHODS
An incidence survey of adults within two hospitals in NHS Scotland was undertaken for one year in 2018/19 as part of the Evaluation of Cost of Nosocomial Infection (ECONI) study. The primary outcome measure was developing any HAI using recognized case definitions. The cohort was derived from routine hospital episode data and linkage to community dispensed prescribing data.


FINDINGS
The risk factors present on admission observed as being the most significant for the acquisition of HAI were: being treated in a teaching hospital, increasing age, comorbidities of cancer, cardiovascular disease, chronic renal failure and diabetes; and emergency admission. Relative risk of developing HAI increased with intensive care unit, high-dependency unit, and surgical specialties, and surgery <30 days before admission and a total length of stay of >30 days in the two years to admission.


CONCLUSION
Targeting patients at risk of HAI from the point of admission maximizes the potential for prevention, especially when extrinsic risk factors are known and managed. This study proposes a new approach to infection prevention and control (IPC), identifying those patients at greatest risk of developing a particular type of HAI who might be potential candidates for personalized IPC interventions.


Introduction
Healthcare-associated infections (HAIs) are a threat to patient safety; they affect patients in all healthcare settings contributing significantly to morbidity and mortality and cost of care [1,2]. Multiple factors contribute to risk of HAI including ageing population, complexity of therapies and emergence of antibiotic-resistant bacteria [3].
HAI describes multiple infection types that can affect any organ; these infections range from urinary tract infections to bloodstream infections (BSIs) [4]. Risk factors for these vary, and as the incidence of HAI at the hospital population level is relatively low, specific infection prevention and control (IPC) measures should be targeted at individuals at increased risk of developing particular HAIs. Identifying and understanding potential risk factors is an essential step in reducing individual risk of developing HAI. Risk factors include intrinsic characteristics of the patient and extrinsic risk factors or exposures that happen to patients during their hospital treatment. Many risk factors have been described for specific populations, for example intensive care unit (ICU) patients' risk of pneumonia [5e14]. However, these studies are by nature retrospective and include risks encountered by patients until the point that they develop HAI. Patient-centred care relies on the identification of high-risk patients on admission in order to enable effective IPC measures to be instigated from the outset. Whereas some studies have focused on extrinsic risk factors and others looked at point-of-transfer risks, some studies report whole-hospital data and a range of HAI types and risks [5,15e18]. Cheng et al. used a model analysing data from electronic patient records in Taiwan, recording intrinsic and extrinsic risk factors throughout the patient stay [18]. They proposed that this could be developed in future to identify patients at risk throughout their stay. This study reports a simpler approach to identify patients at increased risk of HAI on admission. A recent systematic review identified a wide range of intrinsic and extrinsic risk factors and concluded that there was a need to identify patients who are most at risk of HAI, in order to maximize patient safety [19].
This study aimed to assess the relative risk of developing the most frequently occurring HAI types in hospitals. The risk factors included are based on information that would be available on admission and could potentially be used to tailor interventions to manage specific patient risk of HAI. They include a combination of intrinsic and extrinsic factors to which patients have been exposed in the two years before their current admission.

Study design
The Evaluation of Cost of Nosocomial Infection (ECONI) study was an observational incidence surveillance study. The primary outcome measure was HAI defined by the internationally recognized European Centre for Disease Prevention and Control (ECDC) HAI case definitions [4]. Denominator data were collected via routine data sets and linked to the primary data collected on patients who developed HAI. A full list of data items obtained from routine data sets is shown in Appendix Table A2.

Setting
This study was undertaken within a teaching hospital and a general hospital in Scotland for one year ending June 2019. These settings were selected as being broadly representative of other acute hospitals in Scotland in terms of patient specialties, distribution of elective, emergency and transfers, mean length of stay, HAI prevalence, patient mix and rurality. For the two selected hospitals there were 107,244 inpatient admissions in 2014e15 when the study was being planned [20,21]. During the study period the large teaching hospital had 981 beds including 25 intensive care beds (16 general, and nine cardiothoracic), and 33 high-dependency beds (13 general, eight cardiothoracic, and 12 transplant and renal). The large general hospital had 492 beds including five intensive care and four high-dependency beds [22,23].

Participants
All adult overnight inpatient admissions were included within the denominator. Patients admitted as a day case to a study hospital or treated in accident and emergency and discharged without overnight admission were excluded.

Data collection
ECDC HAI epidemiological case definitions were used to define HAI as outcome variable [4]. ECONI study research nurses employed within the participating hospitals reviewed all positive reports from the hospital microbiology databases each day, which highlighted patients with suspected HAI. The patient information systems were then reviewed to establish whether the suspected cases met the ECDC HAI case definitions. These data sources included: patient management systems (TrakCare); prescribing systems; IC (infection control) net; medical and nursing notes; and direct observation. Cases meeting the ECDC HAI case definitions were recorded on a bespoke REDCap database [24,25]. This incidence data set included: hospital, HAI type, date of onset, causative organism, and antibiotic susceptibilities, along with the patients' Community Health Index number (CHI) [22]. CHI is the unique identifier for all patients registered for healthcare in NHS Scotland and was used to link patient records from routine data sources to the denominator record. The CHI number recorded by the ECONI research nurses was used to identify HAI cases within the incidence cohort and all admissions were classified as HAI admissions and non-HAI admissions. HAI admissions contained one or more episodes of HAI. The selection of covariates was based on risk factors identified for developing HAI in a systematic review and meta-analysis [19].

Data items collected through record linkage
The denominator for the incidence cohort study was all hospital stays containing one or more episodes of care in a study hospital within the study period, and was identified from Scottish Morbidity Records 'SMR01'. This data set includes dates and durations of all admissions and discharges to acute hospitals [21].
Data on all eligible inpatients included: date of admission and discharge, specialty, diagnosis for admission (International Classification of Diseases, 10th Revision: ICD-10), surgery defined by Office of Population Censuses and Surveys Classification of Surgical Operations and Procedures (OPCS)-4.8 in two years before admission and during admission, number and location of episodes of care, pre-admission outpatient attendances and surgical interventions in the year before admission [26,27]. The Scottish Index of Multiple Deprivation (SIMD) 2016 quintiles for Scotland were mapped to the National Records of Scotland (NRS) 2019 Scottish Postcode Directory and linked using the SMR01 patient postcode [28]. Charlson Comorbidity Index (CCI) values were calculated using ICD-10 V4 codes from all available SMR01 diagnosis fields, and scores divided into mild (1e2), moderate (3e5), and severe (6) categories for analysis [29,30]. A bespoke specialty group was derived using a combination of significant facility of treatment and specialty of consultant caring for the patients. These groups were high-dependency unit (HDU); intensive care unit (ICU); medicine (MED); obstetrics and gynaecology (OBGYN); all surgical specialties were combined (SUR). A bespoke group of comorbidities was derived based on the ICD-10 codes within SMR dataset. These included diabetes ICD-10 codes (E08eE14); chronic renal failure (CRF) (N17eN19); cardiovascular disease (I00eI99); cancers/malignant neoplasms (C00eC96); chronic pulmonary disease (COPD) (J40eJ47); and nervous system diseases (G00eG99). These all referred to admissions prior to the admission episode analysed in the study.
The number of outpatient hospital visits and any outpatient procedures within the year preceding their admission to hospital were taken from the SMR00 data set, which records outpatient attendances [31]. Linkage to community dispensed prescribing via the 'Prescribing Information System for Scotland' (PIS) provided information on antibiotic exposure in the year before admission [32]. Details of all of the risk factors included in the model are outlined within the Supplementary Appendix. Antibiotic exposure refers to all community dispensed antibiotics and is expressed as the cumulative number of defined daily doses (DDDs) dispensed in the year before admission [33]. Time from when the last antibiotic was taken to admission was calculated. The date corresponding to the end of the antibiotic course was estimated by adding the number of DDDs to the dispensed date, and time to admission calculated in days as the duration from the end of course to admission. Time was set to zero where admission occurred before the estimated last day of taking the antibiotic.

Statistical methods
Each hospital stay contributed once to the analysis. A patient who was readmitted within the study period could contribute more than one stay to the denominator. Admissions with an onset of HAI on or before the date of admission or a date of onset of HAI that did not occur within an inpatient stay were excluded. Episode records within the SMR01 data set were created when a patient was discharged. Patients who remained in hospital at the time of the analysis would not yet have generated a discharge record and were also excluded. Linkage to the routine data sets was undertaken 90 days after the end of the study data collection period to allow for the time lag in data being available for analysis. Records with no valid CHI (N ¼ 10) could not be linked and were excluded. Where a patient's admission contained more than one HAI, the first HAI was used to group the HAI type. The data, sourced from SMR01, were largely complete as these are mandatory data fields that are recorded for all patients [34]. SIMD was linked using patient postcode; where this was missing (N ¼ 340) SIMD was categorized as 'unknown'.
Thirty-two potential risk factors were included within the analysis, and were based on biological plausibility and previously proposed factors within the existing literature [19,20]. This initial set of factors was included within a series of univariate analyses to identify predictors significantly associated with developing HAI. A test for trend was performed for continuous variables by comparing a linear regression model containing the only variable on its own against the null (intercept only) model using likelihood ratio test. No linear trends were identified. Backwards stepwise model selection was performed to identify those risk factors associated with HAI at the 5% significance level. Statistical significance of individual variables was evaluated using Wald's test, and likelihood ratio tests were used to compare nested models in the model build procedure.
A sequential approach was developed when building the model. Risk factors were split into three conceptual themes for modelling to help focus selection: m1, demographics; m2, factors pertaining to the denominator stay; and m3, factors pertaining to the period before admission to the denominator stay. The models were run sequentially, with m2 consisting of those factors identified in the m1 best-fitting model plus additional m2 factors, and m3 consisting of those factors identified in the m2 best-fitting model plus additional m3 factors. Details of the factors included within each model are described in detail in Appendix Table A2. The version retained for multivariate modelling was identified via the lowest Akaike information criterion (AIC) score from the previous best-fitting adjusted model. The factors included within the final model were then used within an adjusted analysis to evaluate the relative risk of the six most frequently occurring HAI types. A multinomial model was run to investigate the relative risk of developing six specific HAI types. All statistical analysis was performed using R software (version 3.5.1) [35].

Ethics
This study was surveillance and therefore was confirmed as ineligible for ethical review (Bailey A. Personal communication to S. Stewart, September 8 th , 2016. South East Scotland Research Ethics Service). It was approved by national information governance approvals: Public Benefit of Health and Social Care: Incidence study: 1617-0037.

Results
During the study period there were a total of 99,018 adult overnight admissions. After the application of statistical restrictions, the analysis included 63,890 eligible admissions to the study hospitals during the study period. There were 44,399 for the teaching hospital and 19,491 for the general hospital ( Figure 1). A total of 893 admissions had one or more HAI, with 822 in the teaching hospital and 71 within the general hospital. Overall incidence was 1.4% with 1.9% within the teaching hospital and 0.4% in the general hospital (reported elsewhere [36]). Table I shows the numbers and percentage of the risk factors included in the final multivariate model within these inpatient stays. Patients aged >60 years made up 61% of admissions. There were similar numbers of males and females. More than half of the admissions had been admitted to hospital at least once in the preceding two years (54.5%) and more than onetenth (12.8%) were patients who had been in hospital for >30 days in the preceding two years. SIMD category two (second most deprived quintile) was the most frequently occurring (26.3%). More than three-quarters of admissions (77.9%) were emergency admissions. Descriptive frequencies and percentages of all potential risk variables included in the analysis are shown in Appendix Table A1. Table I shows the unadjusted and adjusted relative risk (RR) and 95% confidence interval (CI). Comparing within variable groups, patients admitted to a general hospital had significantly lower risk of developing HAI compared with those in the teaching hospital. Risk of developing HAI increased with age, with those aged >80 years showing an RR of 2.47 (CI: 1.82e3.40) compared with those aged <40 years. The highest risk, adjusted for the effects of all other risk factors, was admission to an ICU specialty compared with general medicine (RR: 4.11; 95% CI: 3.34e5.02). This was followed by admission to an HDU specialty (RR: 2.70; 95% CI: 2.08e3.47) compared with a general medical specialty, those with chronic renal failure (RR: 2.19; 95% CI: 1.88e2.54) compared with those without the condition and emergency admissions compared with elective (RR: 2.0; CI: 1.64e2.46). There was an increased risk associated with surgeries within 30 days prior to admission (RR: 1.86; 95% CI: 1.49e2.32), and prior hospital stays of more than two weeks during the previous two years (1.37; 95% CI: 1.08e1.73 for 15e30 days; increasing to RR: 1.65; 95% CI: 1.34e2.04 for >30 days).
Despite showing no association at this point of analysis, sex was retained throughout the analysis as it was considered a clinically important patient characteristic, and later was shown to relate to specific HAI types. Patients in the two least deprived SIMD categories were less likely to develop HAI during their hospital stay than those in the most deprived. Table II shows the adjusted risk by HAI type. There was a reduced risk of onset of any HAI type during a stay in the general hospital compared with the teaching hospital, although the small numbers in the general hospital studied, when analysed by infection type, may reduce the power to accurately detect associations for some infections.
Gastrointestinal infection (GI) also showed an increased risk in those aged >70 years, and in patients aged >80 years the risk was a factor of 5.22 (95% CI: 2.18e12.47) greater than that of those aged <40 years. Emergency admissions showed a 2.36 increase in RR (CI: 1.26e4.43). Patients with diabetes, chronic renal failure, and cancer all showed an increased risk of GI along with those with hospital stays of more than one week in the two years preceding denominator admission. Patients with total length of stay (LOS) >8 days in the two years before admission had an increased RR of GI infection; patients with LOS >30 days had three times greater risk (3.07; 1.79e5.28) than patients with no hospital stay in the previous two years.

Discussion
Identifying patients who are at greater risk of developing HAI at the start of their hospital stay and managing their care to optimize infection prevention is critical. HAI in the hospital population is a relatively infrequent event, with less than 2% of admissions developing HAI during their stay within the ECONI study [36]. This measure of disease frequency is only useful if it can be used to prevent HAI following admission to hospital. Extrinsic risk factors, such as clinical device insertion, are well established in the literature but illustrated only a small proportion of all infections in the ECONI study [36]. Previous studies have tended to focus on specific HAIs, types, causative organisms, or risk factors [19]. However, identifying risk factors that are known at the start of a patient's admission is more valuable because this allows targeted IPC measures to be implemented early [3]. This study provides a novel wholehospital analysis of the factors that increase the risk of HAI, and individual HAI types at the point of admission. The findings have important implications for hospital IPC strategy nationally and locally. This should inform targeting interventions to patients in selected specialties, with the greatly increased risk of developing HAI or a particular HAI type. The opportunity to undertake these types of analyses is rare as data collection is intensive and costly.
The greatest risk factors for development of any HAI were: being treated in a teaching hospital, increasing age, emergency admission and comorbidities of cancer, cardiovascular disease, chronic renal failure, and diabetes. There are differences in teaching and general hospitals in terms of severity of illness of patients and complexity of treatments provided for those patients. Previous studies have shown there to be a higher risk of infection in larger teaching hospitals [37e39]. Comparisons tend to focus on the difference in prevalence of HAI between acute care, combining general and teaching hospitals and primary care combining community and long-stay hospitals [20,40]. Whole-hospital incidence studies are rare and a previous study within the UK focused on a single general hospital [41]. Many of these factors have featured in similar studies to date. However, hospital presentation and socioeconomic status have not been identified in the latest systematic review and require further research [19]. This may be because previous studies have not considered risk for all HAIs at the point of admission, but have retrospectively considered all risks for the entire inpatient stay. Further, the focus of many papers continues to be on those risk factors that are modifiable, rather than considering how best to manage those factors, from an IPC perspective, which cannot be modified.
Hospital type and specialty are an important consideration for prioritizing organizational IPC resources based on risk [42]. In addition, consideration of the historical patient pathway to the current admission is essential. Patients with more than 15 days total stay in hospital, during the two years before their current admission, made up 22.2% of all admissions and those with >30 days were 12.8% of all admissions. Patients at risk of readmissions are linked to long-term conditions and frailty, or emergency admission in younger patients [43]. Previous studies have found that length of stay in older medical patients was related to functional status score, illness severity, cognitive score, poor nutrition, comorbidity score, diagnosis or presenting illness, polypharmacy, age, and gender [44]. With the move from hospital care to managing patients with complex needs in the community, there is clear potential for reduction in both number and length of hospital stay to reduce risk of HAI. Most studies of HAI identify age as a risk factor, in line with the findings here. Older adults are vulnerable to a range of adverse events within the hospital setting compared with younger patients, including malnutrition and pressure ulcers [45e48]. These patients have an increased incidence of comorbidities [49], multiple drugs, and impaired motor and sensory and immune function [50,51]. In line with previous studies, older age increased the relative risk for BSI, GI, and UTI [15,52e54]. The increasing risk of HAI with age, especially in those aged >70 years, points to a need for risk assessment for HAI at point of admission, particularly given the projected 34% increase in the population aged 75 years by 2030 in the UK and what this means for risk of HAI and need for refocused IPC in hospitals [55]. For example, specific interventions, that may appear costly prevention measures if implemented at a wholehospital level, could be targeted to a group of patients who are known to be vulnerable to infection and who have poorer outcomes if they develop HAI.
Multimodal prevention of UTI e the HAI type on which the effect of age is greatest e has been a focus for some years in the form of care bundles for all catheterized patients [56]. There has been a great deal of research into device-related UTI over the last 10 years, especially in relation to antimicrobial materials used in catheters; and although evidence exists for efficacy, there is very limited literature on the costeffectiveness of infection prevention interventions [57,58]. Despite the increased cost of these specialized IPC measures, if overall numbers of infections are reduced significantly and their use is limited to patients who are likely to benefit the most, it is possible that they may well be cost-effective [59].
There is a considerable overlap in the factors predicting extended LOS and risk of HAI, many of which contribute to overall frailty of the patient. Incidence, bed-day use and cost of HAI are often compared with other diseases in the literature, showing that cancers, cardiovascular diseases, and diabetes achieve greater recognition and funding than HAI, when HAI is a similar burden [60]. However, this study has shown that such a 'comparative approach' is a rather simplistic way of looking at the problem [60,61]. The patients with these conditions are the very same patients who are most at risk of HAI.
This study has shown that, for several HAI types, the factors affecting development of HAI that could be identified on admission include some data not currently routinely available to clinical teams. A potential limitation of this risk factor analysis, and of national data sets more broadly, is the exclusion of devices present on admission as a risk factor. However, this is likely to be a small proportion of all admissions to hospital. There is a need for further development of current patient management systems in order to fully achieve this potential. In order to develop the potential for risk assessment on admission, we recommend that hospital admission systems not only include the risk factors identified within this study but in addition record devices in situ. This study demonstrates the potential of creating risk assessment tools on admission, if these data are available, for identifying those most at risk of HAI and prioritizing additional IPC measures in their care pathways to mitigate that risk.
The ECONI surveillance study included all patient admissions to the study hospitals. The two hospitals selected to participate in the study represent about 10% of annual admissions to acute Scottish hospitals. The patient population is similar to the overall NHS Scotland population [21]. Limited identifiable data were recorded for the study; only CHI number was used to divide the cases and non-cases into two groups. This reduced recruitment bias, but resulted in exclusion of those not resident in Scotland (0.1% of all admissions). Charlson comorbidity score data showed that 35.6% of patients did not have a previous admission to derive a CCI score, and that the overall comorbidity summary was not as strong a predictor as a set of individual comorbidities.
In conclusion, this study used a large cohort design to identify and quantify risk of HAI based on factors that are known at point of admission. Most previous studies examined retrospective risk based on cumulative factors that occur during the hospital stay. These findings may help develop a clinical decision tool to identify patients at greatest risk of developing a particular type of HAI. These patients would be potential candidates for personalized infection prevention interventions.