Prognostic models for estimating severity of disease and predicting 30-day mortality of Hypervirulent Klebsiella pneumoniae infections: a bicentric retrospective study

Background Hypervirulent Klebsiella pneumoniae (hvKP) is emerging globally and can cause various, severe infections in healthy individuals. However, the clinical manifestations of hvKP infections are nonspecific, and there is no gold standard for differentiating hvKP strains. Our objective was to develop prognostic models for estimating severity of disease and predicting 30-day all-cause mortality in patients with hvKP infections. Methods We enrolled 116 patients diagnosed with hvKP infections and obtained their demographic and clinical data. Taking septic shock and acute respiratory distress syndrome (ARDS) as the primary outcomes for disease severity and 30-day all-cause mortality as the primary outcome for clinical prognosis, we explored the influencing factors and constructed prognostic models. Results The results showed that increased Acute Physiologic and Chronic Health Evaluation (APACHE) II score [odds ratio (OR) = 1.146; 95% confidence interval (CI), 1.059–1.240], decreased albumin (ALB) level (OR = 0.867; 95% CI, 0.758–0.990), diabetes (OR = 9.591; 95% CI, 1.766–52.075) and high procalcitonin (PCT) level (OR = 1.051; 95%CI, 1.005–1.099) were independent risk factors for septic shock. And increased APACHE II score (OR = 1.254; 95% CI, 1.110–1.147), community-acquired pneumonia (CAP) (OR = 11.880; 95% CI, 2.524–55.923), and extrahepatic lesion involved (OR = 14.718; 95% CI, 1.005–215.502) were independent risk factors for ARDS. Prognostic models were constructed for disease severity with these independent risk factors, and the models were significantly correlated with continuous renal replacement therapy (CRRT) duration, vasopressor duration, mechanical ventilator duration and length of ICU stay. The 30-day all-cause mortality rate in our study was 28.4%. Younger age [hazard ratio (HR) = 0.947; 95% CI, 0.923–0.973)], increased APACHE II score (HR = 1.157; 95% CI, 1.110–1.207), and decreased ALB level (HR = 0.924; 95% CI, 0.869–0.983) were the independent risk factors for 30-day all-cause mortality. A prediction model for 30-day mortality was constructed, which had a good validation effect. Conclusions We developed validated models containing routine clinical parameters for estimating disease severity and predicting 30-day mortality in patients with hvKP infections and confirmed their calibration. The models may assist clinicians in assessing disease severity and estimating the 30-day mortality early.

Currently, the hvKP strain is differentiated from classic Klebsiella pneumoniae (cKP) based on some phenotypic, genotypic properties and determining factors [6].Li G et al. reported that Galleria mellonella killing assay in conjugation with the string test could be used to accurately assess KP virulence and differentiate hvKP from cKP strains [7].Russo TA et al. noted that peg-344, iroB, iucA, prmpA, prmpA2, and siderophore production greater than 30 μg/ml could accurately identify hvKP strains [1,8,9].However, there is no universal standard for identifying all hvKP strains [10].Furthermore, detections of genotype and the determining factors are not widely available, especially in developing countries, making it difficult to recognize hvKP infections early.
Clinical manifestations of hvKP infections, lacking specificity, vary upon the organ involved.Clinically, some patients with hvKP infections soon develop to to septic shock, acute respiratory distress syndrome (ARDS), multiorgan failure and death at final.Early identification of hvKP infections and prediction of disease severity and outcomes are crucial to improve the survival of hvKP-infected patients.Previous studies showed that risk factors for mortality included gastrointestinal fistula, increased Acute Physiology and Chronic Health Evaluation (APACHE) II score and Pitt bacteraemia score, metastatic infection, septic shock, acute respiratory failure and gas formation on imaging [10][11][12].To date, there is few report on risk factors for disease severity.Most of the existing studies of hvKP mainly focus on virulence factors or drug resistance factors at the genetic level, and little attention has been paid to clinical aspects, especially the disease assessment and prognosis models.Therefore, we concentrated on clinical aspects, retrospectively analyzed the demographic and clinical data of hvKP-infected patients to determine the risk factors and tried to construct the prognostic models for disease severity and prognosis.

Study setting and design
Patients with hvKP infections firstly diagnosed at Binhaiwan Central Hospital of Dongguan and Dongguan Tungwah Hospital from September 2017 to September 2022, meeting the inclusion and exclusion criteria, were enrolled in this retrospective study.Demographic and clinical data were collected by two individuals.The protocol for this study was approved by the Medical Ethics Committee of Binhaiwan Central Hospital of Dongguan (No. 2021014).
The exclusion criteria were as follows: 1. Patients younger than 18 years old.2. Patients giving up an active rescue.3. Immunocompromised patients with history of malignancy (under treatment or in remission for less than five years), immunosuppressive disorders (congenital/acquired immunocompromise), use of immunosuppressive regimens (corticosteroid therapy 1 mg/kg/ day prednisone equivalent or corticosteroid therapy for longer than one month, use of another immunosuppressant drug in a high dosage or for longer than one month) [15][16][17].

Detection of virulence-associated features and genes
Hypermucoviscosity was identified by the positive string test.A positive string test was defined as the formation of a viscous string > 5 mm in length when bacterial colonies on an agar plate were stretched with an inoculation loop [3].All KP isolates were stored at − 80 °C until they were sent to relevant institutions (Guangzhou Huayin Health Medical Group Co., Ltd.) for detection of virulence-associated features through targeted nextgeneration sequencing.The genotypic analysis was investigated by polymerase chain reaction with previously described primers [13].High-throughput sequencing was performed using the Illumina MiSeq Reagent Nano Kit.The reads that were correctly aligned at both ends were compared with the reference gene sequence of each virulence gene in the virulence gene data, and finally, the number of reads for each virulence gene in each sample was obtained.
The primary outcomes for the severity of disease were septic shock and ARDS, while the primary outcome for the clinical prognosis was 30-day all-cause mortality.

Statistical analysis
SPSS software (version 13.0) was used for data analysis.Normally and nonnormally distributed continuous variables were summarized as the mean ± standard deviation (SD) and the median with interquartile range (IQR), respectively.Continuous variables were compared using Student's t test or the Mann-Whitney U test, and categorical variables were analyzed by using the χ2 test or Fisher's exact test.P < 0.05 was considered statistically significant.
Univariate and multivariate logistic and Cox regression analyses were used to evaluate the risk factors.Variables with P < 0.05 in the univariate analysis were analyzed in the multivariate model using the likelihood-ratio test.R software (version 4.2.1, CRAN) was used for the nomogram, validation calibration curve, forest plot, scatterplot, receiver operating characteristic (ROC) curve and Kaplan-Meier (K-M) curve.

Clinical features of the patients
A total of 116 patients were enrolled in our study (Fig. 1), their average age was 55.94 ± 15.93 years and 83 (71.6%) patients were male (Table 1).No significant differences were observed in age, gender, history of smoking/alcohol consumption, CAP, or comorbidities (diabetes mellitus, hepatopathy, CKD, cardiovascular disease) between the two hospitals.

Risk factors associated with disease severity
Risk factors associated with septic shock 47/116 (40.5%) patients developed septic shock.The median diagnosis time was 11.63 (4.00, 26.00) hours after admission.There were no significant differences in smoking, alcohol consumption, hepatopathy, CKD or cardiovascular disease between the non-septic shock and septic shock cohorts (P > 0.05).Compared with patients with non-septic shock, the septic shock patients had higher levels of APACHE II score, CRP, PCT, TBIL, DBIL, GLU, CREA, PT and APTT, but lower levels of LYMPH#, MONO#, PLT and ALB.Additionally, there were significantly higher proportions of septic shock patients with CAP, diabetes, bacteremia, extrapulmonary lesion involved, multiple lesions than non-septic shock patients.Regarding clinical outcomes, the septic shock group had longer CRRT duration, vasopressors duration, mechanical ventilator duration and length of ICU stay.(Table 1).
To assess the probability of septic shock, a nomogram with the independent risk factors was constructed, and the calibration curves of the nomogram showed high consistencies between the predicted and actual septic shock probability (Fig. 3A, B).To further investigate the validation of the nomogram, we calculated the septic shock predicted score and drew a correlation analysis scatter plot with CRRT duration, vasopressors duration, mechanical ventilator duration and length of ICU stay respectively.Positive correlations between predicted scores and indices of CRRT, vasopressor, mechanical ventilator and ICU were observed (Fig. 3C-F).

Risk factors associated with ARDS
Seventy-nine (68.1%) patients developed ARDS.The median diagnosis time was 26.00 (18.17, 41.50) hours after admission.No significant difference was found in terms of smoking, alcohol consumption, diabetes, hepatopathy or CKD between the non-ARDS and ARDS groups (P > 0.05).Compared with non-ARDS patients, ARDS patients had higher levels of APACHE II score, GLU, PT and APTT, but lower levels of MONO#, ALB.The ARDS patients had significantly higher proportions of CAP, cardiovascular disease, abscesses, hepatic abscesses and extrahepatic lesion involved and multiple pathogens.For clinical outcomes, the ARDS patients had longer CRRT duration, vasopressors duration, mechanical ventilator duration and length of ICU stay.(Table 1).
To assess the probability of ARDS, a nomogram with the independent risk factors was constructed, and the calibration curves of the nomogram showed high consistencies between the predicted and actual ARDS probability (Fig. 5A, B).To further validate the nomogram, we calculated the ARDS predicted score and drew a correlation analysis scatter plot with CRRT duration, vasopressors duration, mechanical ventilator duration and length of ICU stay.Positive correlations between predicted scores and indices of CRRT, vasopressors, mechanical ventilator and ICU were observed (Fig. 5C-F).

Risk factors associated with 30-day mortality
The 30-day all-cause mortality rate in patients with hvKP infections was 28.4% (33/116).There were no significant differences in the percentages of diabetes, hepatopathy, CKD and cardiovascular disease between the survivors and non-survivors (P > 0.05).Compared with survivors, levels of APACHE II score, CREA, PT and APTT were higher in non-survivors, while the level of ALB was lower.Furthermore, our results revealed that the non-survivors group had significantly higher proportions of smoking, alcohol consumption, CAP, septic shock and ARDS.The non-survivors had longer CRRT duration, vasopressors duration and mechanical ventilator duration, whereas the length of hospital stay was shorter for non-survivors.(Table 2).
We constructed a nomogram of 7-day and 30-day mortality and the calibration curves showed high consistencies between the predicted and actual mortality (Fig. 7A-C).Furthermore, to compare the predictive effects of the prognostic model of 30-day mortality and the APACHE II score, we drew ROC curves of the survival predicted score and APACHE II score.The results showed that the cut-off value of the survival predicted score was 88.765 [area under the curve (AUC) 0.951, specificity 0.792, sensitivity 0.967], and the cut-off value of the APACHE II score was 19.5 (AUC 0.944, sensitivity 0.778, specificity 1.000) (Fig. 7D).We divided the survival predicted score into low-risk group (survival predicted score < 88.765) and high-risk group (survival predicted score ≥ 88.765), and then drew a K-M curve, which showed that the survival rate of the high-risk group was significantly lower than that of the low-risk group (34.1% vs. 98.3%, p < 0.0001) (Fig. 7E).

Discussions
hvKP infections have emerged as a major clinical and public health threat over the past decade [1,10,18,19].As clinical manifestations of hvKP infections are nonspecific, and differentiation of hvKP strains is mainly based on phenotypic and genotypic features without universal standards, it is difficult to identify hvKP infections early.Currently, knowledge of risk factors for disease severity and mortality remains limited.Few studies have investigated the risk factors for mortality in patients with hvKP infections, while no study has explored the risk factors or a prognostic model for disease severity.Since genetic testing is not easy to perform clinically, we summarize the routine clinical parameters to investigate risk factors associated with disease severity and 30-day mortality and construct the prognostic models.This may be the first study to report the risk factors for the severity of hvKP infection, and prognostic models of disease severity and 30-day mortality clinically.Some hvKP-infected patients develop septic shock, ARDS.In our study, the median diagnosis times of septic shock and ARDS were 11.63 (4.00, 26.00) and 26.00 (18.17, 41.50) hours after admission, respectively.Furthermore, patients with septic shock and ARDS had longer CRRT duration, vasopressor duration, mechanical ventilator duration and length of ICU stay.The results suggest that septic shock and ARDS are reasonable predictors for assesssing disease severity in patients with hvKP infections.
Multivariate logistic analysis showed that increased APACHE II score, lower ALB, diabetes, high PCT were independent risk factors for septic shock.Studies on the correlation between APACHE II score and severity of hvKP infections have not been found.A sepsis patient's serum ALB can decrease due to various factors including hypermetabolic state, gastrointestinal dysfunction, capillary leakage [20].There is a causal relationship between hypoalbuminemia and an increased risk of primary and secondary infections, hypoalbuminemia has an effect on the pharmacokinetics and pharmacodynamics of antiinfective drugs, which in turn affects the clinical outcome of infections [21].Hematocrit (HCT)-ALB difference can be a potential predictor for the prognosis of elderly sepsis patients [20].In addition, lower ALB is a risk factor for elderly people with bacterial infections [22], and early infusion of albumin seems to reduce the mortality of  patients with sepsis [23,24].ALB replacement in addition to crystalloids improves the haemodynamics of patients with severe sepsis during the first 7 days [25].Ongoing research on the ALB administration supports the potential for ALB to improve sepsis survival [23].Therefore, it is suggested that correcting hypoalbuminemia possibly reduces the risk of hvKP infections progressing to septic shock, and improves the clinical outcome of hvKP infections.Diabetes mellitus is considered a significant risk factor for acquiring hvKP infections [26][27][28][29], which primarily affects male individuals aged 55-60 years [30].Diabetes is an independent risk factor for KP pyogenic liver abscess [31], as poor glycaemic control might impair neutrophil phagocytosis and promote pathogen growth in tissues, while metabolic disturbances might negatively affect the liver [32,33].Moreover, diabetes, which is more likely to progress hvKP infections, especially hvKP-bloodstream infections (BSIs) [26,28], is an independent risk factor for hvKP-BSIs [12].No studies have been found on the association of PCT with the severity of hvKP infections.However, the PCT level has been shown to be significantly higher in hvKP group compared with cKP group [34].And PCT has been a prognostic biomarker in patients with severe sepsis and septic shock [35].In addition, serum procalcitonin ≥ 5 ng/mL was found to be associated with 30-day mortality of carbapenem-resistant KP infections [36].Our research shows that PCT [19.07(2.95,42.01)ng/mL] is an independent risk factor for septic shock, indicating that PCT is one of the factors predicting the risk of septic shock in patients with hvKP infections.
Multivariate analysis showed that increased APACHE II score, CAP and extrahepatic lesion involved were independent risk factors for ARDS.The APACHE II score is also an independent factor predicting septic shock, so the APACHE II score is very important for the evaluation of hvKP infections.Furthermore, hvKP infections are usually community-acquired [3,37,38], CAP has been showed to be associated with high mortality in patients with hvKP infections [39].Patients with KP pyogenic liver abscesses with sepsis have higher rates of septic shock and acute respiratory distress syndrome [40].
Severe hvKP infections with pyogenic liver abscesses in healthy adults have been reported previously [10,31,37,41].Moreover, liver abscess is a significant risk factor for hvKP infecitons [42], and abscess has been identified as an independent predictor for associated with hvKP-BSIs [43].Nevertheless, our study revealed that hvKP infections with extrahepatic lesion involved were more serious (OR = 14.718), which seems to be inconsistent with previous results.Usually, due to the good permeability of the hepatic sinusoid of the liver, it can promote material exchange between liver cells and blood flow, which is more likely to cause bacteraemia and accelerate the spread of lesions.When the foci of hvKP infections is limited to the liver, which may be related to the immune function of the liver.As a line of defence for immunity, the liver causes a localized lesion and reduces the transfer and dissemination of bacteria to a certain extent, thus reducing the occurrence of bacteraemia and the progression of ARDS.
We constructed prognostic models to assess disease severity, validated the effects of these models, and performed correlation analyses between model scores and clinical outcomes including CRRT duration, vasopressors duration, mechanical ventilator duration and length of ICU stay.Since there were not enough additional cases, only internal validation was performed in this study, and the matching degree of internal validation was good.In the correlation analysis between scores of hvKP infections severity (septic shock, ARDS) and CRRT duration, vasopressors duration, mechanical ventilator duration, and length of ICU stay, the correlation coefficient R (0.43-074) indicated that the correlation was moderately positive.Therefore, septic shock and ARDS are suitable as observation indicators for evaluating the condition of hvKP infections.Due to the small sample size of cases included in this study, further clinical research is needed for verification.
We constructed the prognostic models of 30-day mortality with the variables including age, APACHE II score and ALB level.According to ROC curves of the survival predicted score and APACHE II score, we took survival predicted score = 88.765 as the cut-point, and drew the K-M curves.K-M survival analysis showed that the 30-day mortality of the high-risk group (score ≥ 88.765) was significantly higher than that of the low-risk group (score < 88.765) (34.1% vs. 98.3%, p < 0.0001).The model not only had a good internal validation effect, but also was consistent with previous results.
There were several limitations in our research.Firstly, it was a retrospective study and the sample was quite small.In addition, it was a regional study that all the cases came from Dongguan, which was a labor-intensive city with a large inflow of young people.Finally, external validations of the prognostic models were not feasible due to a lack of additional data.Further investigations are required to confirm these results.

Conclusions
In this retrospective study, increased APACHE II score, decreased ALB, diabetes, higher PCT, CAP and extrahepatic lesion involved were identified as independent risk factors for septic shock and ARDS in patients with hvKP infections.The prognostic models constructed for disease severity with these conventional parameters, were significantly correlated with clinical outcomes, making them potentially practical for clinicians.Moreover, ).E K-M curves of survival predicted score.The survival rate of high-risk group (survival predicted score ≥ 88.765) was significantly lower than that of low-risk group (survival predicted score < 88.765) (34.1% VS 98.3%, p < 0.0001).Abbreviations: APACHE II score, Acute Physiology and Chronic Health Evaluation II score; (ROC) curve, receiver operating characteristic curve; ARDS, acute respiratory distress syndrome.AUC, area under (the) curve; K-M curve, Kaplan-Meier curve younger age, increased APACHE II score, and lower ALB were independent risk factors for 30-day all-cause mortality.The prediction model for 30-day mortality had a good validation effect.In summary, we constructed the prognostic models for disease severity and 30-day mortality in patients with hvKP infections, and the models were helpful for making more practical and effective therapeutic decisions.

Fig. 1
Fig. 1 Flowchart of excluded and included patients.Abbreviations: ARDS, acute respiratory distress syndrome

(Fig. 3
See figure on next page.)The nomogram, calibration curves and correlation analysis scatter plot of septic shock in hvKP infections.A Nomogram with the independent risk factors.B The calibration curves of the nomogram of septic shock (Mean absolute error = 0.049).C Relationships between septic shock predicited score of the nomogram and index of CRRT (CRRT duration/length of hospital stay) (R = 0.44, p < 0.001).D Relationships between septic shock predicited score of the nomogram and index of vasopressors (vasopressors duration/length of hospital stay) (R = 0.71, p < 0.001).E Relationships between septic shock predicited score of the nomogram and index of mechanical ventilator (mechanical ventilator duration/length of hospital stay) (R = 0.44, p < 0.001).F Relationships between septic shock predicited score of the nomogram and index of ICU (length of ICU stay/length of hospital stay) (R = 0.46, p < 0.001).Abbreviations: APACHE II score, Acute Physiology and Chronic Health Evaluation II score; ALB, albumin; CRRT, continuous renal replacement therapy; ICU, intensive care unit

(Fig. 5
See figure on next page.)The nomogram, calibration curves and correlation analysis scatter plot of ARDS in hvKP infections.A Nomogram with the independent risk factors.B The calibration curves of the nomogram of ARDS (Mean absolute error = 0.024).C Relationships between ARDS predicited score of the nomogram and index of CRRT (CRRT duration/length of hospital stay) (R = 0.43 p < 0.001).D Relationships between ARDS predicited score of the nomogram and index of vasopressors (vasopressors duration/length of hospital stay) (R = 0.62, p < 0.001).E Relationships between ARDS predicited score of the nomogram and index of mechanical ventilator (mechanical ventilator duration / length of hospital stay) (R = 0.73, p < 0.001).F Relationships between ARDS predicited score of the nomogram and index of ICU (length of ICU stay/ length of hospital stay) (R = 0.74, p < 0.001).Abbreviations: GLU, glucose; CAP, community acquired pneumonia; ARDS, acute respiratory distress syndrome.CRRT, continuous renal replacement therapy, ICU, intensive care unit time

(Fig. 7
See figure on next page.)The nomogram, calibration curves of assessment models of the 7-day and 30-day all-cause mortality, ROC curve and K-M curve of assessment models of the 30-day all-cause mortality in hvKP infections.A Nomograms of 7-day and 30-day mortality with the independent risk factors.B Calibration curve of the nomogram of 7-day mortality.C Calibration curve of the nomogram of 30-day mortality.The light blue line indicates the ideal reference line where predicted mortality would match the actual mortality.The red dots are calculated by bootstrapping (resample: 1000) and represent the performance of the nomogram.The closer the solid red line is to the light blue line, the more accurately the model predicts mortality.D ROC curves of survival predicted score and APACHE II score.survival predicted score: cutoff value = 88.765(AUC = 0.951 specificity = 0.792, sensitivity = 0.967); APACHE II score: cutoff value = 19.5 (AUC = 0.944, sensitivity = 0.778, specificity = 1.000

Table 2
Clinical variables associated with 30-day mortality of hvKP infections

Table 2
[12,50]ued) 54.15 ± 17.82 years, and younger age was a risk factor for increased mortality, which may be contributed by the violent inflammatory reaction in young people, most of whom showed organ dysfunction, septic shock and ARDS.This point also reminds clinicians that in the face of hvKP infections in young and middleaged adults, modulating the host immune response may be an effective regimen to reduce mortality.In our study, the APACHE II score (HR = 1.157) in the nonsurvivors group was 34.30 ± 9.95.Previous studies have shown that a higher APACHE II score is correlated with a higher 30-day all-cause mortality rate of hvKP infections[12,50], which is consistent with our findings.A low ALB level predicts worse outcomes for patients with BSIs caused by Enterobacteriaceae Fig. 6 Univariate and multivariate cox analyses on variables for the prediction of 30-day all-cause mortality of hvKP infection patients.Abbreviations: HR, Hazard ratio; 95% CI, confidence interval; APACHE II score, Acute Physiology and Chronic Health Evaluation II score; ALB, albumin; CREA, creatinine; PT, plasma prothrombin time; APTT, activated partial thromboplastin time; CAP, community acquired pneumonia; ARDS, acute respiratory distress syndrome was