C-reactive protein predicts complications in community-associated S. aureus bacteraemia: a cohort study

Background Staphylococcus aureus (S. aureus) bacteraemia is increasingly acquired from community settings and is associated with a mortality rate of up to 40% following complications. Identifying risk factors for complicated S. aureus bacteraemia would aid clinicians in targeting patients that benefit from expedited investigations and escalated care. Methods In this prospective observational cohort study, we aimed to identify risk factors associated with a complicated infection in community-onset S. aureus bacteraemia. Potential risk factors were collected from electronic medical records and included: - patient demographics, symptomology, portal of entry, and laboratory results. Results We identified several potential risk factors using univariate analysis. In a multiple logistic regression model, age, haemodialysis, and entry point from a diabetic foot ulcer were all significantly protective against complications. Conversely, an unknown entry point of infection, an entry point from an indwelling medical device, and a C-reactive protein concentration of over 161 mg/L on the day of admission were all significantly associated with complications. Conclusions We conclude that several factors are associated with complications including already conducted laboratory investigations and portal of entry of infection. These factors could aid the triage of at-risk patients for complications of S. aureus bacteraemia.

antibiotic resistance profile of S. aureus [12], whether it persists with treatment [13], and where it is acquired are also possible risk factors for complications [11].
Predictive biomarkers have also been studied. C-reactive protein [CRP] is an acute phase reactant released from the liver and is raised in response to tissue injury and metabolic stressors. Several conditions that increase CRP have been observed including inflammatory processes, cancer, pregnancy and infection [14][15][16].
Although CRP has been rejected as a diagnostic marker for SAB due to the lack of specificity [17], it has also been investigated as a predictive factor of SAB complications. There are limitations to extrapolating these results to a clinical setting. Firstly, international studies on SAB often include healthcare-acquired infections and do not accurately reflect the increasingly common community-associated setting [18]. Other studies do not focus solely upon S. aureus, limiting the translation of the findings [19]. It is important to identify factors, including CRP, that are associated with an increased risk of complicated SAB acquired from the community. This is to better inform management guidelines and impact on mortality rates.
In this study, we aimed to identify potential predicting factors for complication of community-associated SAB infections from all adult cases of SAB that presented to a single institution in Australia. We ascertained several factors that could predict complications and suggest that CRP may be a very useful and readily available biomarker.

Sample selection
University Hospital Geelong [UHG] is a 500-bed tertiary referral hospital in the south-west of Victoria, Australia.
It has a catchment of a population of 600,000 and has a specialised Infectious Diseases service. From June 2015 to September 2018, 236 adult cases of SAB were prospectively identified at UHG. All adult cases admitted to the UHG with an initial positive blood culture of S. aureus flagged within 48 h of admission to hospital were included into the study (Fig. 1). Paediatric cases, culture negative cases, or a blood culture positive after 48 h of admission to hospital [healthcare-acquired] were excluded. Cases not admitted to UHG or transferred to another hospital before a complication was diagnosed were also excluded.

Categorisation
Cases included into the study were categorised on whether they were an uncomplicated or complicated. Categorisation were based upon definitions from the Infectious Disease Society of America Definition [20]. This defines uncomplicated SAB as a SAB which has no infective endocarditis or metastatic infection, no presence of an indwelling medical device within the patient, defervescence within 72 h of antibiotic commencement and no positive blood cultures after 72 h following commencement of antibiotics. Complicated SAB was classified as cases of SAB which did not match the uncomplicated definition.

Data collection
Cases were prospectively identified through the microbiology service (Australian Clinical Laboratories) and clinical data was collected from electronic medical records [EMR]. Data collected included patient demographics, symptomology and portal of entry, microbiological details, and biomarker and blood cell counts from the day of admission. These variables were selected as routinely collected data and their potential as predictive variables for SAB complications. These factors are further detailed below.

Demographics
Demographics included age, gender and co-morbidities. Co-morbidity burden was determined through the Charlson age co-morbidity index [CACI] [21]. The CACI considers age, HIV status, renal and liver function, diabetic status, and peripheral vascular conditions as well as other conditions [21]. Other co-morbidities not used in the CACI, such as injecting drug use, immunosuppressive status, skin conditions and haemodialysis were also collected, as these co-morbidities are frequently associated with S. aureus infections [22][23][24][25][26].

Symptomology and portal of entry
Symptoms were described on admission and recorded in the EMR by the admitting medical officer. These included fever, rigors, nausea, lethargy, and pain. Pain included both local and generalised pain.
Portal of entry was defined as previously described from Smit et al. [27] [27]. We did not include other injections and other entry points as we only had two other entry points observed and no other injections were observed. We instead included community-acquired pneumonia and separated diabetic foot ulcer from SSTI due to the potential increases risk of poor outcomes from these entry points [5,28,29]. SSTIs included pressure ulcers, non-diabetic foot ulcers, cellulitis, and other infections of the skin and soft tissue such as abscesses and boils. Pneumonia was defined as community-acquired pneumonia where there were no other entrance points of infection. IMD included catheters and recent implanted devices. IDU was defined as a SAB infection directly attributable to injecting drug use. An unknown entry point was defined as a case that didn't have any defined entry point.

Laboratory test factors
Microbiological data included whether the bacteraemia was persistent, as defined as having three or more days of consecutive positive blood cultures. The antibiotic sensitivity of the organism was determined through the VITEK 2 system [Biomerieux] and cefoxitin discs and were classified via definitions used by the pathology services. Those pan-sensitive were designated as penicillinsensitive [PSSA], isolates resistant to penicillin were classified as methicillin-sensitive [MSSA], those resistant toward cefoxitin but no other classes were classified as non-multi resistant [nm-MRSA] and those resistant to cefoxitin and other antibiotic classes were classified as methicillin-resistant [MRSA]. It was also noted if the SAB was part of a poly-microbial infection.
Biomarkers and blood tests were ordered at the discretion of the clinical team in charge of the patient on the day of admission. Results of available biomarkers and bloods were collected and included CRP concentration, platelet, leukocyte, neutrophil and lymphocyte cell counts. If a biomarker was collected more than once, the most divergent from the healthy range was chosen.

Statistical analysis Data presentation
The outcome for the study was considered as a binary variable: Complicated infections (Yes/No).The explanatory/predictor variables of interest are both continuous and categorical variables. These include demographics, self-reported symptomology, portal of entry and microbiological and laboratory assessment variables. All categorical data are presented as % [n], and non-normal continuous data are presented as medians and interquartile ranges [IQR].

Univariate analysis-stage 1
Pearson's Chi squared and Fisher's Exact tests and univariable logistic regression were used to determine whether there were associations between the outcome: complicated infections (Yes/No) and potential categorical predictor/exploratory variables. Whereas univariable logistic regression was used to explore potential associations between complicated infections (Yes/No) and continuous exploratory variables. Only results from the univariable logistic regressions are presented.
Receiver operating characteristic (ROC) curve analysisstage 2 Significant blood markers determined from univariate analysis were also assessed for a potential cut-off value.
The main blood marker of interest was CRP concentration. To determine a potential optimal cut-off values, Receiver Operating Characteristic (ROC) curve analysis was used and a cut-off value was predicted using the Youden Index.

Multiple logistic regression-stage 3
Exploratory variables were included in the multiple logistic regression modelling, if preliminary tests had a Wald statistic p-value < 0.2. A stepwise forward and backward elimination approach was used to model certain potential predictors while variables deemed as potential confounders by the literature were included to adjust other associations the multiple logistic regression model. Post estimation analysis such as Hosmer-Lemeshow goodness of fit test, percentage or correctly classified and area under the curve were used to assess the appropriateness of the model. Only patients with complete data were included in the logistic regression modelling. For this exploratory analysis, a level of significance (α) of 5 and 95% confidence intervals were considered appropriate and reported. Although in the discussion of the final adjusted logistic regression model, a more conservative alpha level of 0.01 has been considered. All data were analysed using both Stata SE version 15.1 [30] and SPSS version 24.0 [31].

Description of the cohort
From 2015 until 2018, 236 cases of SAB presented to UHG. After applying exclusion criteria, 158 cases of SAB were included into the study (Fig. 1). Complicated SAB represented 67.6% (n = 107) cases of the study and 32.4% (n = 51) were categorised as uncomplicated.

Portal of entry
An unknown entry point of infection was most common, representing 36.1% (n = 57) of the cohort (Table 3)
Data on blood biomarkers and from full blood examinations collected on the day of admission were analysed and compared (Table 4). Univariable logistic regressions suggested significant associations between complications and the continuous variable of CRP, white blood cell count, and neutrophil cell count. Potentially important cut-off values were determined for each of these variables using Youden's Index and used in further analyses.

Multiple logistic regression analysis
The goodness of fit test suggested that the final multiple logistic regression model shown in Table 5

Discussion
Current reported global mortality rates of SAB range between 10 and 35% [3,5,32,33] and are related to complications of the infection. Identifying complications earlier may improve implementation of appropriate therapy, identify the need for additional interventions and reduce this mortality rate [34]. To that end, several attempts to identify risk factors for complications have been attempted [35][36][37][38]. These studies are varied in definitions, methodology and populations. They also often focus upon one specific complication. Predictive models have also been attempted [39][40][41][42], but focus upon predicting mortality, a specific complication, or are hospital-acquired infections and thus may not be generalizable to the increasingly prevalent communityacquired setting of infection.
To our knowledge, no study has focussed specifically upon community-associated SAB; the predominant cause in Australia. In this study, we aimed to identify risk factors for complications from demographics, self-reported symptoms, portal of entry, and microbiological profile of the bacteria and from routine biomarkers in SAB.
Complicated SAB represented 67.6% of the cohort, with the median age of the cohort 68.5 years (IQR:29.0). Age was significantly associated with complications and this is also reflected in other studies [5,32,36,43].
Haemodialysis (OR adj : 0.09, 95% CI: [0.01, 0.71], p = 0.023) was significantly inversely associated with complications in SAB. This may be due to the increased medical attention received by those who are on haemodialysis. It may be also be due to the portal of entry from an infected line, with early removal associated with improved  Indwelling medical devices relate to any external material and include all catheter types, k-wires, prosthesis and cardiac devices outcomes [44]. However, it must be noted that patients with haemodialysis have been previously associated with increased risk for infective endocarditis [8] in all-cause bacteraemia infections. Significantly, portals of entry were strongly associated with complications in SAB. In this study, an unknown portal of entry was the most common entry point of infection. The higher prevalence of unknown entry point in this study when compared to others, may be due to our definition of portal of entry rather than the infective focus that other studies have used [3,45].
An unknown portal of entry was significantly associated with complications (OR adj : 3.94, 95% CI: [1.33, 11.62], p = 0.013), as seen previously [46]. This may be due the increased risk of persistence due to the inability to identify the portal of entry and remove the infective focus, increasing the likelihood of S. aureus creating metastatic foci.
A portal of entry from an indwelling medical device (OR adj : 4.58, 95% CI: [1.14, 18.35], p = 0.025) was also significantly associated with complications. This may be due to the definition of complication which automatically categorises those with a prosthesis or cardiac device as a complicated case, regardless of whether the medical device is implicated in the infection [20]. Conversely, a portal of entry from a diabetic foot ulcer (OR adj : 0.12, 95% CI: [0.02, 0.77], p = 0.012) was inversely associated with complications. This may be due to increased management of diabetic foot ulcers but could also be due to the suppressed immune effects and reduced blood supply to the ulcer [47].  Investigation into predictive factors from routinely performed blood tests in a clinical setting showed three possible markers associated with complications in univariable analysis. Cut-off values showed high confidence intervals in white blood cell count and neutrophil cell count. This was due to the difference in numbers between the complicated and uncomplicated groups.
It was identified that a CRP concentration of over 161 mg/L on the day of admission was significantly associated with complications (OR adj : 3.56; 95% CI: [1.22, 10.40], p = 0.020). CRP is an acute phase protein of the pentraxin family. It is raised following an inflammatory injury and peaks the following day [48]. It aids in bacterial clearance by activating the classical pathway of the complement system through binding to Cq1 [49]. Interestingly, an inverse relationship has been observed in sepsis cases where a decreased expression of the HLA-DR marker on monocytes was associated with persistently increased CRP concentration over time [19]. CRP has been investigated as a predictive risk factor in adult SAB, with differing outcomes and study populations reflecting all different results. As seen in Table 6, prior studies are varied in terms of setting, population, outcome assessed, and potential cut-off value identified [3.0 mg/L to 850 mg/L]. Our study in comparison, is midrange in terms of study number, has the benefit of being a prospective collection, and attempts to include all complications. Conversely, our study is set in a community-onset SAB population, limiting the generalisability of the findings to all SAB cases.
CRP and full blood examinations are already routine blood immunoassays utilised in the diagnosis of infection and therefore have known parameters and remains stable when affected by delays into processing [60,61]. Therefore, these assays could quickly and efficiently be implemented into a management guideline. CRP could be an additional tool for resource poor healthcare settings to manage the condition by prioritising those with high CRP to have further diagnostic tests to assess for complications. Furthermore, given the increasing positive data on CRP as a predictive marker, it should be highly considered as an accessory marker for triaging patients. Future studies can help confirm the most appropriate cut-off value. It also highlights the potential for other biomarkers to be predictive. Other studies have found several including, procalcitonin, IL-6, IL-10, IL-1β, IL-17, TNFa and IFNy [62][63][64][65][66][67]. However, these markers are not routinely used in a clinical setting and to implement them into clinical practice would require significant additional resources.
There are several limitations to this study. Firstly, this study is from a single centre and the sample of patients is relatively modest. This limits the generalisability of the findings in other settings and potentially a loss of power to detect weaker associations that may exist. However, Pasco and colleagues demonstrated that the Geelong region population closely reflects the Australian population, through comparing the Barwon Statistical Division that represents Geelong and the national average, and showed the most divergent factor [Country of birth] differed by 9.5% [67]. Nonetheless, results of the study are promising, and we plan to extend this work to a larger study with multiple sites to further validate our model.

Conclusion
In this study, we aimed to identify potential predictive factors of complications in adult community-associated SAB. This included a quantified cut-off predictive marker of any potential blood markers already used in the clinical setting. It was identified that a history of haemodialysis was protective against complications in this cohort. The importance of the portal of entry in the outcome of SAB was observed. An unknown entry point and an entry point originating from an IMD were both significantly associated with complications, whereas a portal of entry from a DFU was protective against complications. Most importantly, we identified that a CRP over 161 mg/L on the day of admission was significantly associated with complicated disease and might offer a useful predictive tool to aid clinicians in the management of adult cases of community-associated SAB.