非人类免疫缺陷病毒感染的肺孢子菌肺炎患者早期预后模型的建立

Establishment of a multiple predictor model of non-HIV pneumocystis carinii pneumonia

目的探讨非人类免疫缺陷病毒(HIV)感染的肺孢子菌肺炎(PCP)患者的预后影响因素,建立早期多指标联合预测模型。方法回顾性分析北京协和医院2015年3月至2018年3月收治的非HIV感染的PCP患者的临床资料155例,采用单因素分析和Logistic回归分析,筛选影响非HIV感染的PCP患者预后的独立因素,绘制各独立预后因素和预后预测模型的受试者工作(ROC)曲线,确定临界值、ROC曲线下面积(AUC)、敏感度、特异度、阳性预测值和阴性预测值,观察各独立预测因素和预后预测模型的预测效果。结果 Logistic回归模型的结果表示,肾脏疾病、氧合指数(Pa O2/Fi O2)、乳酸脱氢酶(LDH)和细菌感染是非HIV感染的PCP预后的独立影响因素(肾脏基础病:χ^(2)=11.268、P=0.001,PaO_(2)/FiO_(2):Z=-4.413、P<0.000,LDH:Z=4.013、P<0.001,细菌感染:χ^(2)=5.082、P=0.034,P均<0.05),各指标的AUC分别为0.6075、0.6874、0.6973、0.5913;敏感度分别为0.880、0.622、0.742、0.640,特异度分别为0.350、0.701、0.575、0.500;阳性预测值分别为0.559、0.667、0.613、0.550,阴性预测值分别为0.757、0.659、0.712、0.597。最终预测模型为-0.7835×A-0.00528×B+0.00197×C+0.4269×D(基础病为肾脏疾病A赋值为1,基础病为非肾脏疾病A赋值为0;B为Pa O2/Fi O2的值;C为LDH的值;合并细菌感染D为1,无细菌感染D为0),当该值> 1时死亡风险增加,该模型的ROC曲线AUC为0.7636,敏感度、特异度、阳性预测值、阴性预测值分别为0.769、0.620、0.649、0.746。结论肾脏疾病、Pa O2/FiO_(2)、LDH和合并细菌感染是影响非HIV感染的PCP患者预后的独立因素,该早期多指标预后预测模型的临床预测效果良好,可为PCP患者早期预后评估提供参考。

Objective To explore the prognostic factors of the patients with non-human immunodeficiency virus( HIV) pneumocystis carinii pneumonia( PCP) and to establish an early multiple predictor model. Methods Clinical data of 155 non-HIV-PCP patients who met the inclusion criteria between March 2015 and March 2018 at Peking Union Medical College Hospital were retrospectively collected and analyzed. The independent factors related to the prognosis of non-HIV-PCP patients were screened by univariate analysis and Logistic regression analysis. Combined with various independent prognostic factors,an early multiple predictor model was constructed,and the receiver operation characteristic( ROC) curves of the independent prognostic factors and the predictor model were produced. The cutoff value of independent prognostic factors and predictor models were determined. The area under the curve( AUC),the sensitivity,the specificity,positive predictive value( PPV) and negative predictive value( NPV) were calculated. Results The results of Logistic regressionmodel showed that basic renal diseases( χ^(2)= 11. 2682,P = 0. 001),PaO_(2)/FiO_(2)( Z =-4. 413,P<0. 0001),LDH( Z = 4. 0133,P< 0. 0001) and bacterial infection( χ^(2)= 5. 082,P = 0. 034) were independent prognostic factors( P< 0. 05). Respectively,the AUC of each index was 0. 6075,0. 6874,0. 6973,0. 5913,and the sensitivity was 0. 880,0. 622,0. 742,0. 640,the specificity was 0. 350,0. 701,0. 575,0. 500,PPV were 0. 559,0. 667,0. 613,0. 550,NPV were 0. 757,0. 659,0. 712,0. 597. The final prediction model =-0. 7835 × A-0. 00528 × B + 0. 00197 × C + 0. 4269 × D( If the basic disease was kidney disease,A was assigned as 1,and 0 for non-renal basic disease;B was the value of PaO_(2)/FiO_(2);C was the value of LDH;D was defined as 1 with bacterial infection or 0 without bacterial infection). When the value of the model was larger than 1,the risk of death increased. The AUC of the model was 0. 7636,and the sensitivity,specificity,PPV and NPV were 0. 769,0. 620,0. 649 and 0. 746,respectively. Conclusions Renal diseases,PaO_(2)/FiO_(2),LDH and bacterial infection are independent factors affecting the prognosis of non-HIV-PCP patients. The early multi-index predictive model has a good clinical predictive effect and can provide the reference for the early evaluation of PCP.