Enhancement of CURB65 score with proadrenomedullin (CURB65-A) for outcome prediction in lower respiratory tract infections: Derivation of a clinical algorithm

Lower respiratory tract infections (LRTIs) comprising community-acquired pneumonia (CAP), exacerbation of chronic obstructive pulmonary disease (COPD) and acute bronchitis, are the most important infectious causes of death in industrialized countries [1]. In the management of CAP, assessment of disease severity and prediction of outcome is essential for a rational allocation of health care resources and for guiding therapeutic options. For this purpose, different risk assessment tools have been developed. The CURB65 is a widely used risk assessment tool [2, 3]. It relies on only five predictors, is relatively easy to memorize and calculate, and demonstrates better practicability than other scores [4, 5]. However, it conceptually is a static score and lacks information on the host's inflammatory response. The score was validated to predict mortality only and has not been proven to predict other CAP-associated adverse outcomes. CURB65 has been validated in many studies for CAP patients, and to a lesser extent in other non-CAP-LRTI patients [6, 7]. It provides only limited discriminatory power with moderate sensitivity and specificity. This potentially leads to hospitalization of low-risk patients where outpatient treatment would be preferable, and conversely, to outpatient treatment of high-risk patients [4]. Thus, there is increasing interest in new risk factors and biomarkers that confer additional prognostic information [8, 9].

Several candidate hormones have been proposed, among which proadrenomedullin (ProADM) proved to be the most promising in patients with sepsis [10–12], CAP [13–15] and other LRTIs [13, 16]. ProADM belongs to the calcitonin peptide superfamily and emerges from the CALC V gene with distinct molecular regulation and various effects, including potent vasodilatation, immune modulation and bactericidal activity [16, 17]. ProADM is a "hormokine", characterized by a hormone-like behavior in non-inflammatory conditions when it is produced only by endocrine cells; and a cytokine-like behavior in septic conditions when it is ubiquitously hyper-expressed [18, 19]. We recently validated the prognostic performance of ProADM in a large cohort of LRTI patients from a multicenter study [13]. A precise clinical algorithm for future routine use, however, remains to be established. The aim of this analysis was, therefore, to derive a risk score based on CURB65 classes enhanced by ProADM cut-off ranges for improved prognostic assessment in CAP and non-CAP-LRTIs as a basis for future intervention studies.

Combining CURB65 classes with the prognostic information of ProADM in a new risk classification (CURB65-A score) offers improved risk prediction with regard to both adverse events and mortality in CAP and non-CAP-LRTI patients. This could lead to reclassification of alleged low-risk patients based on the CURB65 classification into higher risk classes and consecutive inpatient treatment; at the same time, almost 20% of patients who were treated as inpatients would be considered low risk based on CURB65-A and outpatient treatment would be strongly recommended. Thus, the results of his study suggest that the CURB65-A score can efficiently and safely increase the proportion of outpatient treatment in all LRTI patients, CAP and non-CAP.

Both PSI and CURB65 are recommended by guidelines for site of care decision in patients with CAP [29]. However, both scores were initially developed to predict mortality [3, 5, 30], while daily decisions have to take into account other adverse outcomes, in addition to comorbidities and non-medical factors. In the ProHOSP study, fear of adverse medical outcome was one of the most important perceived reasons by healthcare providers leading to hospital admission. This was true independent of medical risks as assessed with clinical risk scores [31]. In the ProHOSP study, 8.7% patients with CURB65 score 0-1 had an adverse event and 0.9% died, which may be considered unacceptably high by many physicians and patients to recommend or accept outpatient treatment. Therefore, in reality, triage decisions in CAP patients are not always based on these clinical risk scores [32] and are not even evidence- or guideline-based for non-CAP-LRTI. Obvious advantages of outpatient treatment include a reduced risk of health-care associated infections [33] and other nosocomial complications such as falls or delirium of elderly patients [34]. Outpatient treatment is preferred by most patients [35], improves patient and relatives' satisfaction [36], and would importantly free scarce resources in many healthcare settings, which face considerable bed shortages during the respiratory season [37]. In addition, inpatient treatment for LRTI is 8-20 times more expensive than outpatient treatment [24, 36, 38]. Marrie et al. [39] demonstrated that 13.5% of CAP patients with high risk based on a PSI level of IV or V, could be safely treated as outpatients with only 0.6% of those requiring ultimate hospital admission and a 0.6% mortality.

Thus, modified and improved triage algorithms are necessary [40], and new clinical pathways have been proposed [41, 42]. Both of these studies were performed in Canada and little data is available on the generalizability and routine applicability of these complex pathways. Yealy et al. [43] used graded-intensity level measures to improve guideline utilization in a trial-setting in US emergency departments. However, the necessary efforts raise doubt about the feasibility in daily routine.

In contrast, readily available biomarkers which are easy to measure and quantify, might provide objective and dynamic measurements supporting clinical judgment. In addition, such biomarkers may provide an added level of confidence for physicians to actually follow recommendations of a scoring system. For prognostic purposes, ProADM is superior to PCT and other biomarkers as shown by our group and others [10, 11, 13, 15]. These findings were largely confirmed in a multicenter US study enrolling 1653 patients with CAP, even though in the US study ProADM only added prognostic value in patients with severe CAP (PSI classes IV&V, CURB65 groups 2&3), while mortality in less severe CAP (PSI classes I-III, CURB65 group 1) was low irrespectively of ProADM values [14]. Our data confirm that ProADM significantly improves the CURB65 score for prediction of adverse events and in mortality in the high CURB65 classes. Similarly, in our study patients with low CURB65 classes (0-1) had low mortality with ProADM tertiles providing no additional information quoad vitam. However, increasing ProADM tertiles were associated with significantly increased risk of adverse events even in the low risk CURB65 classes.

Taken together, combining ProADM values with CURB65 scores allows a more objective triage of patients with LRTI presenting to the emergency department. It improved the prediction of adverse events in low CURB65 classes which might increase physicians' compliance with triage decisions and their acceptance by patients. Adherence to the combined risk score CURB65-A would also improve the safety of patients who formerly would have been discharged based on a low CURB65 score but in fact carried an unacceptably high risk of complications. Overall, the CURB65-A has great potential to increase the number of secure outpatient treatments by improved triage confidence and compliance.

Another strength of this study is its design as a pre-planned substudy of the randomized controlled multicenter ProHOSP intervention trial. Since it involved patients in large and medium-sized academic and non-academic hospitals, the results will be largely generalizable to many settings. This study provides evidence that CURB65 performed similarly for outcome prediction in CAP and non-CAP-LRTI patients, as recently reported [6], importantly extending their clinical utility.

There are limitations of our study. In line with the original ProHOSP study [20], we used a composite of adverse events defined as all-cause mortality, ICU admission, or any disease specific complications (i.e. empyema in all cases) as our primary endpoint. While this was done to incorporate clinically meaningful outcomes in light of sample size considerations, there are disadvantages of composite endpoints [44], such as differences in the importance of each part of the composite. One may argue that while all-cause mortality is an objective endpoint, ICU admission depends in part on the experience of the physician in charge, critical-care bed availability and other "soft" factors; however, we displayed severity criteria within ProHOSP in order to standardize ICU admission. In addition, all adverse events were monitored by an independent data safety and monitoring board. Still, because a high CURB65 score was one of the severity criteria which prompted physicians to consider admitting patients to the ICU, its prognostic performance may be artificially improved. As a consequence, this would imply a rather conservative bias in regard of the prognostic performance of ProADM. In analogy to existing risk scores and guideline recommendations, we only included medical risk factors. However, for real-life triage decisions, social, societal, organizational, functional factors and preferences of patients and relatives also have to be taken into account. In some settings these non-medical factors might be even more important and lead to reduced effectiveness of the algorithm. We are currently planning an intervention study with a multidisciplinary risk assessment including these factors to assess the efficacy of the CURB65-A score and the real-life effectiveness of the interdisciplinary triage algorithm. Finally, since the number of patients was higher in the CAP than in the non-CAP-LRTI group and since our primary endpoint was mainly driven by the outcomes in the CAP group with relatively low mortality and adverse event rates in the non-CAP-LRTI group, the utility of the CURB65-A score for patients with non-CAP-LRTIs is less certain. The similar results of the algorithm in patients with and without CAP, however, are reassuring. Nonetheless, this heterogeneous but clinically important group of non-CAP-LRTI patients deserves particular attention in future studies testing the CURB65-A score.

In conclusion, the combination of ProADM and CURB65 in a novel risk score (CURB65-A) showed improved performance for triaging patients with regard to expected risks for mortality and adverse outcomes compared to CURB65 alone. Whether the theoretical benefits will actually translate into real-life improvements with improved safety by hospitalizing patients at high risk and a greater effectiveness by recommending ambulatory treatment for patients with LRTI at low risk has to be answered by future intervention studies.