Publication: Desarrollo y validación de un modelo predictivo del riesgo de infección basado en parámetros inmunológicos en receptores de trasplante renal
Loading...
Official URL
Full text at PDC
Publication Date
2018-11-27
Authors
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Complutense de Madrid
Citations
Exportar
Abstract
La infección constituye una de las principales causas de morbilidad y mortalidad entre receptores de trasplante renal (TR). Sin embargo, nuestra capacidad en el momento actual para identificar a los pacientes en riesgo de esta complicación es limitada. El desarrollo de un modelo ponderado de predicción basado en parámetros inmunológicos no patógeno-específicos y en variables clínicas permitiría individualizar el seguimiento y abordaje de estos pacientes en función de su riesgo específico de infección. Métodos: Análisis retrospectivo de una cohorte observacional prospectiva integrada por 447 pacientes sometidos a TR en nuestro centro entre noviembre de 2008 y marzo de 2013. Esta muestra global fue segmentada de forma aleatoria en una cohorte de derivación (292 pacientes) y una cohorte de validación (155 pacientes). Monitorizamos el recuento de subpoblaciones linfocitarias (linfocitos T CD3+, CD4+ y CD8+, linfocitos By linfocitos natural killer [NK]) y los niveles sé ricos de inmunoglobulinas y factores de complemento (C3 y C4) según un esquema preestablecido: momento basal (pre-trasplante), mes 1 y mes 6. El objetivo del estudio fue el desarrollo de infección global y bacteriana a lo largo de diversos periodos post-trasplante (precoz [primer mes], intermedio [meses 1 a 6) y tardío [a partir del mes 6]). Para la construcción de los modelos se realizaron varios análisis de regresión logística (según el periodo y tipo de infección), asignando un valor numérico a los coeficientes B así obtenidos. La capacidad de discriminación fue analizada mediante el área bajo la curva de características operativas del receptor (ROC); la exactitud diagnóstica mediante la estimación de sensibilidad, especificidad, y val ores predictivos positivo (VPP) y negativo (VPN); y la reproducibilidad y calibración mediante la cohorte independiente de validación. Resultados: No hubo diferencias en los parámetros inmunológicos basales en función del desarrollo de infección durante el primer mes post-trasplante, por lo que no se pudo construir ningún modelo para este periodo. El modelo de predicción de infección global durante el periodo intermedio se basó en los siguientes parámetros (evaluados al mes 1): edad del receptor ≥ 63 años (4 puntos), reintervención en el primer mes (3 puntos), función del injerto renal (-0,3 puntos por cada intervalo de 10 mL/ min), linfocitos T CD4+ <30 céls/mcL (3 puntos), linfocitos B <110 céls/mcL (2 puntos), lgA <115 mg/dL (3 puntos) y C3 <76 mg/dL (2 puntos). El área bajo la curva ROC fue 0,771. Empleando como punto de corte >4 puntos observamos los siguientes valores: sensibilidad 54,8%, especificidad 83,2%, VPP 60,7% y VPN 79,5%. El modelo para infección bacteriana consistió, por su parte, en los siguientes parámetros: edad del receptor ≥ 65 años (3 puntos), retraso en la función del injerto (2 puntos), infección bacteria na previa (4 puntos), linfocitos NK <68 céls/mcL (2 puntos) e lgG <500 mg/dL (3 puntos). El área bajo la curva ROC fue 0,765. Con un punto de corte >6,5 puntos obtuvimos: sensibilidad 54,4%, especificidad 85,4%, VPP 47,7% y VPN 88,4%. El modelo de predicción de infección global durante el periodo tardío incluyó los siguientes parámetros (evaluados al mes 6): enfermedad aterotrombótica (3 puntos), rechazo agudo previo (2 puntos), linfocitos NK <232 céls/ mcL (6 puntos) y C3 <103 mg/dL (2 puntos). El área bajo la curva ROC de 0,736. Empleando un punto de corte >7 puntos observamos los siguientes valores: sensibilidad 69,2%, especificidad 64,3%, VPP 41,4% y VPN 85,2%. Por último, el modelo para infección bacteriana tardía consistió en: infección por VHC (3 puntos), rechazo agudo previo (3 puntos) y C3 <103 mg/dL (3 puntos). El área bajo la curva ROC fue de 0,687. En este caso, el umbral óptimo (>4,5 puntos) arrojó los siguientes valores: sensibilidad 37,8%, especificidad 90,3%, VPP 45,2% y VPN 87,3%. La reproducibilidad y calibración de los modelos en la cohorte de validación fue buena. No hubo diferencias significativas en las áreas bajo la curva ROC ni en las incidencias de infección tras estratificar ambas cohortes por cuartiles o terciles del correspondiente modelo de predicción. Conclusiones: La combinación de parámetros inmunológicos no patógeno-específicos y ampliamente disponibles junto con una serie de variables clínicas seleccionadas permite predecir, en los meses 1 y 6 post-trasplante, el riesgo posterior de infección global o bacteriana en receptores de TR de forma individualizada.
Infection still constitutes one leading cause of morbidity and mortality among kidney transplant (KT) recipients. However, our current capacity to identify those patients at high risk for such event is limited. The development of a weighted prediction score based on a number of non-pathogen-specific immune parameters and clinical variables may be useful to individualize post-transplant management according to the predicted risk. Methods: We performed a retrospective analysis of a prospective observational cohort including 447 patients that underwent KT at our center from November 2008 to March 2013. This overall sample was randomly split into a derivation cohort (292 patients) and a second validation cohort (155 patients). We assessed peripheral blood lymphocyte subpopulations (CD3+, CD4+ and CD8+ T-cells, B-cells and natural killer [NK] cells) counts and se rum immunoglobulin and complement (C3 and C4) levels according to a pre-established schedule: baseline (pre-transplant) and post-transplant months 1 and 6. The study outcome was the occurrence of overall and bacterial infection throughout different post-transplant periods (early [first month], intermediate [months 1-6] and late [>6 months]). We constructed a set of logistic regression models (according to each post-transplant period and the type of infection predicted) and assigned a numeric value to the resulting B coefficients in order to developed the prediction scores. The discriminative power was assessed by means of the area under receiver operating characteristic (auROC); diagnostic accuracy was analyzed by sensitivity, specificity, positive (PPV) and negative predictive (NPV) values; and reproducibility and calibration by means of the independent validation cohort. Results: We found no differences in baseline immune parameters according to the occurrence of infection during the first post-transplant month, and therefore no score for the early period could be developed. The model for predicting overall infection during the intermediate period was based on the following variables (assessed at month 1): recipient age ≥ 63 years (4 points), reintervention within the first month (3 points), kidney graft f unction (-0.3 points per 10 ml/min-interval), CD4+ T-cell count <30 cells/ mcl (3 points), B-cell count <110 cells/mcl (2 points), lgA levels <115 mg/dl (3 points) and C3 levels <76 mg/dl (2 points). The auROC was 0.771. By applying the optimal cut off value (>4 points) we obtained the following diagnostic accuracy: sensitivity 54.8%, specificity 83.2%, PPV 60.7% and NPV 79.5%. The model for predicting bacterial infection during this period included the following parameters: recipient age <:65 years (3 points), delayed graft function (2 points), previous bacterial infection (4 points), NK cell count <68 cells/mcl (2 points) and lgG level <500 mg/dl (3 points). The corresponding auROC was 0.765. The optimal cut-off value (>6.5) yielded: sensitivity 54.4%, specificity 85.4%, PPV 47. 7% and NPV 88.4%. The model for the prediction of overall infection during the late post-transplant period was constructed on the basis of the following parameters (assessed at month 6): pre-transplant atherothrombotic disease (3 points), previous graft rejection (2 points), N K cell count <232 cells/mcl (6 points) and C3 level <103 mg/ dl (2 points). The auROC was0.736. With the optimal cut-off value (>7 points) we observed: sensitivity 69.2%, specificity 64.3%, PPV 41.4% and NPV 85.2%. Finally, the model for late bacterial infection included chronic HCV infection (3 points), previous graft rejection (3 points) and C3 level <103 mg/dl (3 points). The au ROC was 0.687. The diagnostic accuracy by applying the optimal cut-off value (>4,5) was: sensitivity 37.8%, specificity 90.3%, PPV 45.2% and NPV 87.3%. Model reproducibility and calibration within the validation cohort were good. The were no significant differences in the auROCs between the derivation and validation cohort, or in the incidence of infection after stratifying both cohorts by increasing quartiles or tertiles of each prediction model. Conclusions: The combination into a single predict ion model of easily available non-pathogen-specific immune parameters and a series of selected clinical variables is useful to predict, at post-transplant months 1 and 6, the individual risk of subsequent infect ion in KT recipient.
Infection still constitutes one leading cause of morbidity and mortality among kidney transplant (KT) recipients. However, our current capacity to identify those patients at high risk for such event is limited. The development of a weighted prediction score based on a number of non-pathogen-specific immune parameters and clinical variables may be useful to individualize post-transplant management according to the predicted risk. Methods: We performed a retrospective analysis of a prospective observational cohort including 447 patients that underwent KT at our center from November 2008 to March 2013. This overall sample was randomly split into a derivation cohort (292 patients) and a second validation cohort (155 patients). We assessed peripheral blood lymphocyte subpopulations (CD3+, CD4+ and CD8+ T-cells, B-cells and natural killer [NK] cells) counts and se rum immunoglobulin and complement (C3 and C4) levels according to a pre-established schedule: baseline (pre-transplant) and post-transplant months 1 and 6. The study outcome was the occurrence of overall and bacterial infection throughout different post-transplant periods (early [first month], intermediate [months 1-6] and late [>6 months]). We constructed a set of logistic regression models (according to each post-transplant period and the type of infection predicted) and assigned a numeric value to the resulting B coefficients in order to developed the prediction scores. The discriminative power was assessed by means of the area under receiver operating characteristic (auROC); diagnostic accuracy was analyzed by sensitivity, specificity, positive (PPV) and negative predictive (NPV) values; and reproducibility and calibration by means of the independent validation cohort. Results: We found no differences in baseline immune parameters according to the occurrence of infection during the first post-transplant month, and therefore no score for the early period could be developed. The model for predicting overall infection during the intermediate period was based on the following variables (assessed at month 1): recipient age ≥ 63 years (4 points), reintervention within the first month (3 points), kidney graft f unction (-0.3 points per 10 ml/min-interval), CD4+ T-cell count <30 cells/ mcl (3 points), B-cell count <110 cells/mcl (2 points), lgA levels <115 mg/dl (3 points) and C3 levels <76 mg/dl (2 points). The auROC was 0.771. By applying the optimal cut off value (>4 points) we obtained the following diagnostic accuracy: sensitivity 54.8%, specificity 83.2%, PPV 60.7% and NPV 79.5%. The model for predicting bacterial infection during this period included the following parameters: recipient age <:65 years (3 points), delayed graft function (2 points), previous bacterial infection (4 points), NK cell count <68 cells/mcl (2 points) and lgG level <500 mg/dl (3 points). The corresponding auROC was 0.765. The optimal cut-off value (>6.5) yielded: sensitivity 54.4%, specificity 85.4%, PPV 47. 7% and NPV 88.4%. The model for the prediction of overall infection during the late post-transplant period was constructed on the basis of the following parameters (assessed at month 6): pre-transplant atherothrombotic disease (3 points), previous graft rejection (2 points), N K cell count <232 cells/mcl (6 points) and C3 level <103 mg/ dl (2 points). The auROC was0.736. With the optimal cut-off value (>7 points) we observed: sensitivity 69.2%, specificity 64.3%, PPV 41.4% and NPV 85.2%. Finally, the model for late bacterial infection included chronic HCV infection (3 points), previous graft rejection (3 points) and C3 level <103 mg/dl (3 points). The au ROC was 0.687. The diagnostic accuracy by applying the optimal cut-off value (>4,5) was: sensitivity 37.8%, specificity 90.3%, PPV 45.2% and NPV 87.3%. Model reproducibility and calibration within the validation cohort were good. The were no significant differences in the auROCs between the derivation and validation cohort, or in the incidence of infection after stratifying both cohorts by increasing quartiles or tertiles of each prediction model. Conclusions: The combination into a single predict ion model of easily available non-pathogen-specific immune parameters and a series of selected clinical variables is useful to predict, at post-transplant months 1 and 6, the individual risk of subsequent infect ion in KT recipient.
Description
Tesis de la Universidad Complutense de Madrid, Facultad de Medicina, Departamento de Medicina, leída el 14/03/2018