The aim of our study was to assess the performance of the CAP severity score CURSI  in predicting short-term mortality and to compare it to the established scores CRB-65 and CURB-65. Our findings show that increasing score points and risk categories of CURSI are associated with increasing 30-day mortality in CAP patients. However, CURSI does not outperform the other scores, and has some limitations.
One of the most important steps in managing CAP involves deciding whether a patient can safely be regarded as being at low risk of mortality. This would offer the perspective to align several management decisions to risk, with hospital admission being the most important decision. Admission to hospital multiplies the treatment costs, harbours the danger of hospital-acquired infections and is associated with a slower return to work and usual activities [19–22]. Therefore, our analysis focused on the ability of the severity scores to identify patients at low risk of mortality, or in other words on the sensitivity of the severity scores used. In comparison to CRB-65 and CURB-65 (sensitivity about 95% and 85%, respectively), CURSI had a markedly lower sensitivity (around 42%). The 95% CI showed no overlap between CRB-65 or CURB-65 and CURSI. In our view, this finding indicates a clinically important difference in sensitivity between CBR-65 or CURB-65 and CURSI that is unlikely to be explained by chance alone. The mentioned finding is also reflected by the negative likelihood ratios found. The negative likelihood ratios show that being identified as low risk by CRB-65 or CURB-65 indicates a much lower chance for mortality at day 30 than being defined as low risk by CURSI.
Compared to CURB-65, CURSI identifies twice as many patients as being at low risk. However, this possible advantage is nullified by the high mortality in this group. This limitation could only be avoided by lowering the proposed cut-off . The resulting values (sensitivity, specificity, number of patients and mortality in the low-risk group) would be very similar to those of CURB-65. However, a modification of the CURSI-defined low-risk group would require further validation studies. By contrast, CRB-65 and CURB-65 have already been well validated in recent years [9, 11–15].
Two further aspects deserve consideration. The risk categories as proposed for the CURSI do only differentiate between a low risk and a high risk group. By contrast, CRB-65 and CURB-65 based categories also define an intermediate risk group. Current guidelines [23, 24] do not recommend basing the indication for ICU admission on severity scores, like CRB-65 or CURB-65. However, the differentiation of patients not being low risk may allow a further differentiation of treatment intensities, for example intensity of clinical monitoring. Further, in our opinion, replacing the criteria age and blood pressure of CURB-65 by shock index in CURSI offers no practical advantage. Age can be assessed easily in nearly all patients. Blood pressure is also part of the shock index and thus has to be measured in both cases.
The motivation for developing CURSI was the assumed inaccuracy of the criteria age and blood pressure of CURB-65 with reference to the high prevalence of hypertension in elderly patients and the low correlation between chronological and biological age . The shock index was purported to be a better marker for pneumonia severity and had already been shown to correlate with mortality in CAP patients . We can confirm this observation with our data. However, we can also confirm that age is a strong predictor of short-term mortality, which has already been shown repeatedly [15, 25–27]. On this basis, a simple, yet likely explanation why CURSI is not able to outperform CRB-65 or CURB-65 would be that it only substitutes one strong predictor variable (age) with another (shock index).
The test performance of CURB-65 and CRB-65 found in our analysis is comparable to other studies. In their analysis describing and evaluating the CURB-65, Lim et al.  reported a sensitivity of 93% and a specificity of 49% for the identification of the low risk group. The performance of the CRB-65 was worse (77% and 64% sensitivity and specificity, respectively), but still comparable. Likewise, several other studies found sensitivity and specificity for CURB-65 to be between 80%-90% and 50%-60%, respectively [9, 13, 28]. In the original CURSI study , Myint et al. reported only 60% and 75% for sensitivity and specificity for the CURB-65. For CURSI, the figures were 61% and 72%, which is somewhat better than the estimates we derived from our analysis. However, as it seems, the equivalent results of Myint et al. for CURSI and CURB-65 are caused by the relatively weak performance of the CURB-65 rather than a good performance of the CURSI.
The AUROC of the CRB-65 was lower compared to the other severity scores. This can be explained by the omission of the urea criterion, which has been shown to be significantly associated with mortality in CAP patients [7, 29–31]. Furthermore, CRB-65 identified fewer patients at low risk than CURB-65 (87 vs. 211), but showed better sensitivity with a resulting even slightly lower mortality in the low-risk group. These findings are consistent with several other studies [7, 9, 12, 28, 32]. CRB-65 is recommended for use in ambulatory care [6, 23, 24], as CURB-65 requires the blood urea value, which may not be available in all cases or might delay decision-making. In a large prospective German study including inpatients and outpatients with CAP , blood urea was available for fewer than half of the outpatients included in the study.
Our study has limitations. One limitation is the retrospective design and the use of registry data for analysis. However, we do not believe that this affects the validity of our findings. The registry was established for the purpose of quality assurance and contains reliable and complete information on all variables of the CRB-65. Data are entered into the database in time, i.e. during hospital stay of the patient or immediately after discharge. The corresponding blood urea values could be retrieved from the central laboratory server without any missings. Pulse rate was clearly documented on admission in all but six cases (1.1% of all cases). Reliable information on 30-day mortality as the primary endpoint was provided for all patients by the relevant local register office. With this, we are convinced that the data for the variables of interest, namely the three severity scores used and mortality as the primary endpoint, are valid. Another limitation is that only inpatients were included. Therefore, no statement can be made as to whether patients identified as being at low risk can be safely treated as outpatients. This hypothesis demands testing in prospective, randomised controlled trials. Finally, we present data from a single centre only. In terms of CRB-65 and CURB-65 performance, our results are in line with several other studies. However, as characteristics of hospitals and populations served can differ widely, generalisability of our findings is certainly limited.