Clinical Features Of Pulmonary Mycosis And Establishment Of Risk Prediction Model For Pulmonary Candidiasis In Diabetic Patients

Objective Diabetic patients are prone to infections,particularly pulmonary infection.The burden of diabetes mellitus(DM)is especially heavy in China due to the large population and high prevalence of diabetes.Previous studies have shown that DM is a risk factor for incidence of invasive pulmonary mycosis(IPM),the diagnosis of which is difficult at early stage associated with poor outcome and high mortality.Currently,large sample size clinical study specifically addressing pulmonary mycosis in diabetic patients is still lacking.This study analyzed the pathogen distribution and clinical characteristics of IPM in diabetic patients using multi-center retrospective case-control design;constructed risk prediction models for pulmonary candidiasis in diabetic patients using the common clinical data;and evaluated the performance of the established model in clinical practice to provide a quick and reliable objective tool for clinical decision-making.Methods Part ⅠA total of 157 cases of IPM in DM patients were retrospectively identified from the inpatients in Shanghai Changhai Hospital and Zhongshan Hospital.The clinical data were collected from all the patients,including demographic data,clinical characteristics,serological tests,radiological findings,diabetes-related parameters,and causative pathogens.IPM patients were compared to analyze the clinical features of different fungal pathogens so as to provide preliminary evidence for differentiating different types of IPM at early stage.Part ⅡAccording to the predefined searching strategy and inclusion/exclusion criteria,255 DM patients were identified with pulmonary infection and positive culture of Candida spp from respiratory tract(92 cases of IPM and 163 cases of Candida colonization).The clinical data of these patients were collected for constructing the risk prediction models for IPM in diabetic patients using Logistic regression method and random forest(RF)algorithm,respectively.The two established models were preliminarily evaluated for their performance in clinical practice.Results PART Ⅰ: Clinical characteristics of IPM in DM patients1.Fungal pathogens and diagnostic certainty of IPM in DM patients A total of 157 cases of IPM in DM patients were identified.The fungal pathogens included Candida in 92 cases(proven diagnosis in 6 cases/clinical diagnosis in 86 cases),Aspergillus in 34 cases(proven diagnosis in 4 cases/clinical diagnosis in 30 cases),Cryptococcus in 26 cases(proven diagnosis in 18 cases/clinical diagnosis in 8 cases),Pneumocystis in 4 cases(all clinical diagnosis),and Mucor in 1 case(proven diagnosis).2.IPM features of different fungal pathogens(1)General data The patients were assigned to different groups according to their fungal pathogens.The mean age of patients was 69.9 ± 13.1 years in Candida group,62.1 ± 11.0 years in Aspergillus group,and 58.9 ± 10.0 years in Cryptococcus group.The patients in Candida group were significantly older than those in Aspergillus and Cryptococcus groups(both P<0.001).More patients in Candida group had underlying hypertension(70.0%)compared to Aspergillus group(35.3%,P < 0.001)and Cryptococcus group(42.3%,P = 0.011).Significantly more patients in Candida group had underlying cerebrovascular disease(30.4%)than in Aspergillus group(5.9%,P = 0.004)and Cryptococcus group(5.9%,P =0.001).Coronary heart disease was found in 32.6% of the patients in Candida group,significantly higher than in Cryptococcus group(3.9%,P = 0.003).Renal insufficiency was found in 20.7% of the patients in Candida group,significantly higher than in Cryptococcus group(0.0%,P = 0.011).Significantly more patients in Aspergillus group had chronic obstructive pulmonary disease than in Candida group(23.5% vs 7.6%,P =0.014).Pneumonia Severity Index(PSI)was used to evaluate the severity of the disease.Cryptococcus group had lower PSI score than Candida group [69.5(60.5-82.5)vs 114.5(98.8-139.0)points] and Aspergillus group [69.5(60.5-82.5)vs 100.5(84.3-123.0)points].2.Host factors No significant difference was observed in host factors between Candida group and Aspergillus group.Significantly more patients in Candida group(39.1%)and Aspergillus group(47.1%)used broad-spectrum antibiotics before onset of infection than Cryptococcus group(3.9%,P < 0.001).Generally,41.3% of the patients in Candida group and 26.5% of the patients in Aspergillus group received central venous catheterization;43.5% of the patients in Candida group and 23.5% of the patients in Aspergillus group received mechanical ventilation.The patients in Cryptococcus group did not receive central venous catheterization or mechanical ventilation.3.Diagnosis certainty The rate of proven diagnosis was the highest in Cryptococcus(69.2%),significantly higher than that in Candida group(11.8%,P < 0.001)and Aspergillus group(6.5%,P <0.001).4.Main symptoms Fever was reported in significantly lower percentage of patients in Cryptococcus group(19.2%)than in Candida group(52.2%,P = 0.003)and Aspergillus group(64.7%,P< 0.001).Sputum was recorded in significantly lower percentage of patients in Cryptococcus group than in Candida group(19.2% vs 47.8%,P = 0.009).Chest tightness was documented in significantly lower percentage of patients in Cryptococcus group than in Aspergillus group(7.7% vs 35.3%,P = 0.012).5.Imaging features The most common lesion type was patchy infiltration in Candida group(81.5%),significantly higher than in Aspergillus group(50%,P<0.001)and Cryptococcus group(15.4%,P < 0.001).The most common lesion type was nodular mass in Cryptococcus group(80.8%),significantly higher than in Aspergillus group(23.5%,P = 0.001)and Cryptococcus group(7.6%,P<0.001).Cavity lesion was more common in Aspergillus group than in Candida group(23.5% vs 6.5%,P = 0.007).Bilateral lesion was found in83.7% of the patients in Candida group and 61.8% of the patients in Aspergillus group,however the lesion was mostly unilateral in Cryptococcus group(57.7%).The lesion distribution was significantly different between Candida group and Cryptococcus group(P<0.001).Pleural effusion was more common in Candida group(35.9%,P<0.001)and Aspergillus group(29.4%,P = 0.002)than in Cryptococcus group(0.0%).6.Laboratory tests1)G-test: Aspergillus group showed higher G-test results than Candida group [53.0(21.8-126.2)vs 20.6(10.4-72.1),P = 0.015].2)Infection-related biomarkers: WBC in Cryptococcus group was lower [7.1(5.4-9.0)] than in Candida group [10.5(7.6-15.2)] and Aspergillus group [9.0(6.4-12.1)].The percentage of neutrophils in Cryptococcus group[68.2%(64.1%-71.7%)] was lower than in Candida group [83.8%(77.8%-89.3%)] and Aspergillus group [84.5%(70.0%-90.2%)].C-reactive protein(CRP)in Cryptococcus group [5.6(1.1-12.0)] was significantly lower than in Candida group [69.1(13.7-98.3)]and Aspergillus group [38.5(16.6-112.0)].3)Other tests: The hemoglobin(128.7 ± 19.9)and albumin [42.0(39.3-43.0)] in Cryptococcus group were significantly higher than in Candida group [(109.9 ± 25.3),31.0(28.0-34.0)] and Aspergillus group [(109.1±24.1),31.0(28.0-35.5),all P < 0.001].7.Diabetes related factors1)The fasting blood glucose in Cryptococcus group [8.3(6.6-11.5)] was lower than in Candida group [11.7(8.9-15.8),P = 0.002] and Aspergillus group [12.5(9.0-15.8),P =0.016].2)The glycosylated hemoglobin in Cryptococcus group [7.2(6.7-8.1)] was lower than in Candida group [8.2(7.1-9.6),P = 0.01] and Aspergillus group [8.6(7.9-9.4),P =0.002].PART Ⅱ Establishment and validation of risk prediction model for respiratory tract Candida culture-positive diabetic patients with pulmonary infection progressing to pulmonary candidiasis1.A total of 255 diabetic patients complicated with pulmonary infection and culture positive of Candida spp.from respiratory tract were included in this study,including 92 cases of IPM and 163 cases of Candida colonization in respiratory tract.The baseline characteristics were similar between IPM cases and colonization cases except significant difference in age,complication of connective tissue disease,and cavity lesion by imaging study.These 3 factors were included in Logistic univariate regression analysis.The total days of hospital stay [18.5(13.0-30.0)vs 11.0(7.0-16.5),P<0.001] and ICU stay [0.0(0.0-10.0)vs 0.0(0.0-0.0),P < 0.001] were significantly longer in IPM group than in control group.2.Univariate analysis was performed between groups using diabetes-related variables,peripheral blood tests,PSI score,and host factors as independent variables.The results showed that insulin use,fasting blood glucose,glycosylated hemoglobin,albumin,neutrophil percentage,G test,PSI score and risk class,chemotherapy,corticosteroid therapy,mechanical ventilation,broad-spectrum antibiotic use,and hemodialysis were significantly different between the two groups.3.Binary Logistic univariate analysis was conducted with the 18 diabetes-related and other significant variables.The results suggested that connective tissue disease,cavity lesion,insulin use,fasting blood glucose,Hb A1 c,Hb A1 c grade,hemoglobin,serum albumin,chemotherapy,hemodialysis,systemic corticosteroid therapy,mechanical ventilation,broad-spectrum antibiotic therapy,central venous catheterization,PSI score,PSI risk class were associated with the progression of Candida colonization to pulmonary candidiasis.4.The significant outcome-related variables identified in the Logistic univariate analysis were included for stepwise AIC regression.Hb A1 c grade,systemic corticosteroid therapy,mechanical ventilation,broad-spectrum antibiotic use,central venous catheterization,and PSI risk class(P < 0.05)were selected as predictors to establish the risk prediction model for progression of Candida colonization to pulmonary candidiasis(Model 1).The nomogram for the model was plotted.The area under the receiver operating characteristic(ROC)curve(AUC)was 0.854.This model showed good discrimination.The calibration curve of the nomogram for Model 1 demonstrated good agreement between prediction and observation.Decision Curve Analysis(DCA)demonstrated that patients could benefit from the clinical decision based on this prediction model.Results of computer resampling(Bootstrap)for 500 times implied that the AUC under ROC curve was 0.855(95% CI 0.813-0.905),suggesting the outstanding prediction performance of this model.5.A total of 33 variables(including gender,age,underlying diseases,imaging characteristics,diabetes related variables,peripheral blood tests,and host factors)were included.Seven variables including broad-spectrum antibiotic treatment,central venous catheterization,mechanical ventilation,Hb A1 c,systemic corticosteroid therapy,chemotherapy and PSI risk class and were screened out by RF algorithm through sorting and dimensionality reduction.The risk prediction model for pulmonary candidiasis in diabetic patients with Candida colonization in respiratory tract was established in the training set and verified the test set internally.The AUC under ROC curve in the test set was 0.815.The sensitivity was 0.792,and the specificity was 0.857.The model had good prediction performance.Conclusions1.The clinical cases of IPM in diabetic patients were reviewed in this study.It was found that Candida,Aspergillus,and Cryptococcus were the most common fungal pathogens of IPM in diabetic patients.2.The clinical characteristics of IPM in DM patients varied with the specific fungal pathogen.The patients with pulmonary cryptococcosis generally had better condition than the patients with pulmonary aspergillosis or candidiasis.Most of pulmonary cryptococcosis patients were confirmed cases,with relative fewer underlying diseases,better blood glucose control,fewer fever and respiratory symptoms,and more unilateral nodules and masses by imaging study.The patients with pulmonary candidiasis were older,with longer diabetes duration,more underlying hypertension and cerebrovascular disease,more severe disease,more symptoms,more bilateral lesions of patchy infiltration by imaging study,and more frequent pleural effusion.The patients with pulmonary aspergillosis was more frequently complicated with COPD,with chest tightness symptoms,and more cavity lesion by imaging study.3.Target population of the prediction model established in this study is diabetic patients,complicated with CAP,who failed to respond to adequate antimicrobial therapy,and were Candida positive in respiratory specimens.4.This study developed a Nomogram scoring tool,which is based on the readily available clinical parameters.The tool is useful in predicting the probability of progression of Candida colonization in respiratory tract to pulmonary candidiasis in diabetic patients.The discrimination,calibration,and clinical significance of the model demonstrated that this prediction model scoring tool could be used for assessment of individual patients before relevant examination and antifungal therapy.This tool is helpful for improving the early diagnosis of pulmonary candidiasis and guiding clinical decisionmaking.5.This study also explored the utility of RF algorithm in the construction of prediction model with small sample data preliminarily.RF algorithm is promising in model construction compared with the model established by Logistic regression method.