Patient characteristics
A total of 1318 patients with positive blood culture was identified. Nosocomial infections, n = 352, and blood cultures positive with bacteria regarded as contamination, n = 375, were excluded. SOFA-score was incomplete in 129 patients. The missing information to calculate complete SOFA-scores were coagulation-score points n = 28 patients, liver-score points (n = 126) and kidney-score points (n = 2). Out of these 129, 93 reached ≥ 2 points despite incomplete SOFA-score and were therefore included. The remaining patients, n = 36, were considered non-valid and excluded because of lack of information to classify SOFA-score. Evaluation of the patients’ medical records and assessment of the exclusion criteria yielded 555 patients included for further analyses. For further details, see Fig. 1.
Median age was 76 with 51.7% males and CCI was distributed with 0 points in 28.8%, 1–2 points in 40.7% and ≥ 3 points in 30.5% of the patients. Gram-positive bacteria were found in 45.0% and no patients had multibacterial findings. Most common pathogens were Escherichia coli 35.1%, followed by Staphylococcus aureus 15.7% and Streptococcus pneumoniae 13.0%. (Table 1)
|
Valid
(n = 555)
|
Non-valid*
(n = 36)
|
P***
|
Total
(n = 591)
|
Table 1
Baseline characteristics among patients with positive blood culture
Age, median (IQR, min-max)
|
76 (65 to 85, 18–98)
|
76 (61 to 86, 21–94)
|
0.84
|
76 (65 to 85, 18–98)
|
Women, n (%)
|
268 (48.3%)
|
22 (61.1%)
|
0.14
|
290 (49.1%)
|
Men, n(%)
|
287 (51.7%)
|
14 (38.9%)
|
|
301 (50.9%)
|
CCI, n (%)**
|
|
|
< 0.01
|
|
0
|
160 (28.8)
|
21 (58.3)
|
|
181 (30.6)
|
1–2
|
226 (40.7)
|
9 (25.0)
|
|
235 (39.8)
|
3+
|
169 (30.5)
|
6 (16.7)
|
|
175 (29.6)
|
Gram pos, n (%)
|
250 (45.0)
|
16 (44.4)
|
0.94
|
266 (45.0)
|
S. aureus, n (%)
|
87 (15.7)
|
6 (16.7)
|
0.87
|
93 (15.7)
|
S. pneumoniae, n (%)
|
72 (13.0)
|
4 (11.1)
|
> 0.99
|
76 (12.9)
|
Gram neg, n (%)
|
305 (55.0)
|
20 (55.6)
|
|
325 (45.0)
|
E. coli, n (%)
|
195 (35.1)
|
17 (47.2)
|
0.14
|
212 (35.9)
|
Others, n (%)
|
201 (36.2)
|
9 (25.0)
|
0.17
|
210 (35.5)
|
*incomplete SOFA-score and therefore not possible to classify. For further details, see Table S2 |
**CCI – Charlson Comorbidity Index |
***Mann-Whitney test for age and chi-2 test or Fischer exact test if appropriate for the other variables.
The frequency of all pathogens are presented in detail in supplementary data, Table S1.
The presence of sepsis according to SOFA-score was 425 of 555, 76.6% and 28 days mortality was 67 of 555, 12.1%. The non-valid classified group (n = 36) had statistically significant lower CCI score (p < 0.01) compared to the valid population. (Table 1)
The ability of NEWS2 to detect sepsis
The distribution of NEWS2 score is shown in Fig. 2.
425 patients (76.6%) had SOFA score ≥ 2 point and classified as sepsis. 431 (77.6%) had NEWS2 score ≥ 5 points and considered as positive. The sensitivity for NEWS2 to detect sepsis was 87% (95% CI 0.83–0.90. Kruskal-Wallis test shows heterogeneous groups with significantly higher NEWS2-score in patients with infections caused by Streptococcus pneumoniae, P = 0,014, but with no other significant findings.
The sensitivity of NEWS2 to detect sepsis was 87% (95% CI 0.83–0.90). The sensitivity was 85% for Gram-negative infections (OR = one reference) and 89% for Gram-positive infections ORadj 1.62 (95% CI 0.89–2.93), P = 0.11. The corresponding sensitivity for specific pathogens was for Escherichia coli 85% (OR = one reference), 85% for Staphylococcus aureus ORadj 1.03 (95% CI 0.44–2.44) and 92% for Streptococcus pneumoniae ORadj 2.59 (95% CI 0.89–7.55). The ability of NEWS2 to predict sepsis was not statistically associated with the etiology of the infection. The specificity of NEWS2 detection of sepsis was low, 52% (95% CI 0.43–0.59) with no statistically significant differences observed between pathogens. (Table 2)
Table 2
Diagnostic accuracy of NEWS2 classification compared to SOFA-score in total and by bacterial species
a Positive NEWS2-score was determined if the sum of all individual points were ≥ 5 points. b Adjusted for age in 5 years intervals (< 65, 65–69, 70–74, 75–79, 80–84, 85–89, ≥ 90), sex and CCI (0, 1–2, ≥ 3). PPV Positive predicted value; NPV Negative predicted value; CI confidence interval
Out of the 555 included patients, 431 had a NEWS2 ≥ 5 and the PPV to detect sepsis was 85% (95% CI 0.82–0.88). In the group of patients with Gram-positive etiology, the PPV for detecting Sepsis was 89%. The corresponding figure in patients with Gram-negative etiology was 82% (REF.), ORadj 1.97 (95% CI 1.09–3.56), P = 0.02. This difference was no longer significant when non-valid cases were included in the sensitivity analyses and interpreted as a dilution of the results (Table S1). PPV to detect sepsis in patients with Streptococcus pneumoniae was 92%, ORadj 3.44 (95% CI 1.18-10.0), P = 0.02. The corresponding numbers for Staphylococcus aureus was 88%, ORadj 1.40 (95% CI 0.56–3.49), P = 0.47. (Fig. 2)
The NPV of NEWS-2 for excluding Sepsis was low, 48% (95% CI 0.45–0.63) with no statistically significant differences observed between different species (Table 2). To adjust for possible effects of excluding the non-valid cases, a sensitivity analyses was performed. In this analyses non-valid cases were included (n = 36) and labeled as SOFA negative since they did not reach 2 points. The same calculations as in Table 2 were performed (sensitivity, specificity, PPV and NPV). The only differences observed with this analysis was that Gram positive bacteria no longer was significant in PPV to detect sepsis. For further details, please see Table S1 in supplementary data.