Uso del enfoque bayesiano para la estimación y proyección de tasas / Use of a bayesian approach to rate estimation and projection

OBJETIVO: Aplicar y valorar el enfoque bayesiano para realizar proyecciones de tasas de mortalidad por cáncer a través del ajuste de modelos edad-período-cohorte (EPC). MÉTODOS: El método de estimación bayesiano se aplica a datos de mortalidad por cáncer de vejiga en Argentina. Se adopta un esquema autorregresivo de segundo orden para la especificación a priori de los coeficientes del modelo EPC. Se comparan las estimaciones obtenidas con toda la información disponible y excluyendo los grupos de edad con tasas de mortalidad bajas, a fin de valorar el comportamiento del enfoque ante datos esparcidos. Se proyectan las tasas de mortalidad a dos períodos sucesivos a los observados. RESULTADOS: Se comprueba la robustez del método, lo cual evita excluir los grupos de edad con tasas de mortalidad nulas o bajas. Las tasas observadas caen todas dentro de las bandas de credibilidad y confirman la bondad del ajuste del modelo. Se observa una tendencia general decreciente de las tasas de mortalidad por cáncer de vejiga. Las estimaciones y proyecciones de estas tasas resultan más precisas en los grupos etarios que presentan mayor incidencia de mortalidad. CONCLUSIONES: La formulación bayesiana utilizada permite reducir la variación aleatoria entre estimaciones adyacentes al especificar que los efectos de cada escala dependan de los inmediatos anteriores. Se demuestra la capacidad del enfoque para manejar frecuencias bajas y obtener estimaciones confiables de las tasas de mortalidad, como así también proyecciones precisas sin necesidad de realizar supuestos adicionales como sucede en el ajuste clásico de un modelo EPC.

OBJECTIVE: Apply and assess a Bayesian approach to projecting cancer mortality rates by fitting age-period-cohort (APC) models. METHODS: The Bayesian estimation method was applied to bladder cancer mortality data in Argentina. A second-order autoregressive model was adopted for a priori specification of APC model coefficients. The estimates obtained were compared with all available information and excluding age groups with low mortality, to assess behavior of the approach in light of scattered data. Mortality was projected for two successive periods following the ones observed. RESULTS: Robustness of the method was verified, which avoids excluding age groups with null or low mortality. Observed rates all fall within the credibility bands and confirm the model's goodness of fit. An overall downward trend in bladder cancer mortality was observed. Estimates and projections of these rates are more precise in age groups that have greater incidence of mortality. CONCLUSIONS: The Bayesian formulation used herein makes it possible to reduce random variation between adjacent estimates by specifying that the effects of each scale depend on the immediately preceding ones. It was demonstrated that the approach has the capacity to handle low frequencies and obtain reliable mortality estimates, as well as precise projections, without the need for making additional assumptions, as happens in classical APC model fitting.

Uso del enfoque bayesiano para la estimación y proyección de tasas / Use of a Bayesian approach to rate estimation and projection

Objetivo. Aplicar y valorar el enfoque bayesiano para realizar proyecciones de tasas de mortalidad por cáncer a través del ajuste de modelos edad-período-cohorte (EPC). Métodos. El método de estimación bayesiano se aplica a datos de mortalidad por cáncer de vejiga en Argentina. Se adopta un esquema autorregresivo de segundo orden para la especificación a priori de los coeficientes del modelo EPC. Se comparan las estimaciones obtenidas con toda la información disponible y excluyendo los grupos de edad con tasas de mortalidad bajas, a fin de valorar el comportamiento del enfoque ante datos esparcidos. Se proyectan las tasas de mortalidad a dos períodos sucesivos a los observados. Resultados. Se comprueba la robustez del método, lo cual evita excluir los grupos de edad con tasas de mortalidad nulas o bajas. Las tasas observadas caen todas dentro de las bandas de credibilidad y confirman la bondad del ajuste del modelo. Se observa una tendencia general decreciente de las tasas de mortalidad por cáncer de vejiga. Las estimaciones y proyecciones de estas tasas resultan más precisas en los grupos etarios que presentan mayor incidencia de mortalidad. Conclusiones. La formulación bayesiana utilizada permite reducir la variación aleatoria entre estimaciones adyacentes al especificar que los efectos de cada escala dependan de los inmediatos anteriores. Se demuestra la capacidad del enfoque para manejar frecuencias bajas y obtener estimaciones confiables de las tasas de mortalidad, como así también proyecciones precisas sin necesidad de realizar supuestos adicionales como sucede en el ajuste clásico de un modelo EPC.

Objective. Apply and assess a Bayesian approach to projecting cancer mortality rates by fitting age-period-cohort (APC) models. Methods. The Bayesian estimation method was applied to bladder cancer mortality data in Argentina. A second-order autoregressive model was adopted for a priori specification of APC model coefficients. The estimates obtained were compared with all available information and excluding age groups with low mortality, to assess behavior of the approach in light of scattered data. Mortality was projected for two successive periods following the ones observed. Results. Robustness of the method was verified, which avoids excluding age groups with null or low mortality. Observed rates all fall within the credibility bands and confirm the model’s goodness of fit. An overall downward trend in bladder cancer mortality was observed. Estimates and projections of these rates are more precise in age groups that have greater incidence of mortality. Conclusions. The Bayesian formulation used herein makes it possible to reduce random variation between adjacent estimates by specifying that the effects of each scale depend on the immediately preceding ones. It was demonstrated that the approach has the capacity to handle low frequencies and obtain reliable mortality estimates, as well as precise projections, without the need for making additional assumptions, as happens in classical APC model fitting.

[Use of a bayesian approach to rate estimation and projection]. / Uso del enfoque bayesiano para la estimación y proyección de tasas.

OBJECTIVE: Apply and assess a Bayesian approach to projecting cancer mortality rates by fitting age-period-cohort (APC) models. METHODS: The Bayesian estimation method was applied to bladder cancer mortality data in Argentina. A second-order autoregressive model was adopted for a priori specification of APC model coefficients. The estimates obtained were compared with all available information and excluding age groups with low mortality, to assess behavior of the approach in light of scattered data. Mortality was projected for two successive periods following the ones observed. RESULTS: Robustness of the method was verified, which avoids excluding age groups with null or low mortality. Observed rates all fall within the credibility bands and confirm the model's goodness of fit. An overall downward trend in bladder cancer mortality was observed. Estimates and projections of these rates are more precise in age groups that have greater incidence of mortality. CONCLUSIONS: The Bayesian formulation used herein makes it possible to reduce random variation between adjacent estimates by specifying that the effects of each scale depend on the immediately preceding ones. It was demonstrated that the approach has the capacity to handle low frequencies and obtain reliable mortality estimates, as well as precise projections, without the need for making additional assumptions, as happens in classical APC model fitting.

[Application of the intrinsic estimator to breast cancer mortality rates]. / Aplicación del estimador intrínseco a tasas de mortalidad por cáncer de mama.

OBJECTIVE: Assess use of the intrinsic estimator (IE) technique in epidemiology. METHODS: The IE approach was applied to the analysis of breast cancer data in Argentina in order to observe the trends associated with "age, period, and cohort" (APC). This method involves the use of a principal components regression to obtain a single set of estimated trends. The results were compared to the findings obtained with the conventional method, which consists of adjusting a generalized linear model that includes the traditional constraints of the statistical model as well as an additional constraint (CGLM). RESULTS: Both methods yielded compatible results in the trends associated with APC. However, they differed in the confidence intervals, with IE yielding greater efficiency. The curve associated with age showed the expected pattern of change across the life course: the greater the age, the greater the risk. With regard to cohorts, a decrease in the effects associated with the most recent cohorts was evident, whereas there was very little variation in the estimated effects for the period. CONCLUSIONS: A comparison of the results obtained with the IE method and the CGLM method revealed the reach of the generic solution provided by the IE to the problem of estimates in an APC model. The IE method is based on conversion of the data observed using a weighting matrix that is simple to apply and provides estimates with desirable statistical properties.

Aplicación del estimador intrínseco a tasas de mortalidad por cáncer de mama / Application of the intrinsic estimator to breast cancer mortality rates

OBJETIVO: Valorar la utilización de la técnica del estimador intrínseco (EI) en el campo de la epidemiología. MÉTODOS: Se aplicó el enfoque EI al análisis de datos sobre cáncer de mama en la Argentina a fin de observar las tendencias asociadas a "edad, período y cohorte" (EPC). Esta metodología recurre a la aplicación de una regresión por componentes principales para obtener un único conjunto de tendencias estimadas. Se compararon sus resultados con los obtenidos por el método convencional "modelos lineales generalizados restringidos" (MLGR) que incluye una restricción adicional a las tradicionales en el modelo estadístico. RESULTADOS: Ambos métodos proporcionaron resultados compatibles en las tendencias asociadas a EPC, aunque difirieron en los intervalos de confianza, con mayor eficiencia por parte del EI. La curva asociada a edad mostró el patrón de cambio esperado según el transcurso de la vida: a mayor edad, mayor riesgo. En relación a cohortes, se evidenció un decrecimiento de los efectos asociados a las cohortes más recientes, mientras que para período los efectos estimados presentaron muy poca variación. CONCLUSIONES: La comparación entre los resultados obtenidos por el método EI y el de MLGR reveló los alcances de la solución genérica provista por el EI al problema de estimación en un modelo EPC. El método EI se basa en una transformación de los datos observados utilizando una matriz de ponderaciones de sencilla aplicación y proporciona estimaciones con propiedades estadísticas deseables.

OBJECTIVE: Assess use of the intrinsic estimator (IE) technique in epidemiology. METHODS: The IE approach was applied to the analysis of breast cancer data in Argentina in order to observe the trends associated with "age, period, and cohort" (APC). This method involves the use of a principal components regression to obtain a single set of estimated trends. The results were compared to the findings obtained with the conventional method, which consists of adjusting a generalized linear model that includes the traditional constraints of the statistical model as well as an additional constraint (CGLM). RESULTS: Both methods yielded compatible results in the trends associated with APC. However, they differed in the confidence intervals, with IE yielding greater efficiency. The curve associated with age showed the expected pattern of change across the life course: the greater the age, the greater the risk. With regard to cohorts, a decrease in the effects associated with the most recent cohorts was evident, whereas there was very little variation in the estimated effects for the period. CONCLUSIONS: A comparison of the results obtained with the IE method and the CGLM method revealed the reach of the generic solution provided by the IE to the problem of estimates in an APC model. The IE method is based on conversion of the data observed using a weighting matrix that is simple to apply and provides estimates with desirable statistical properties.

[High relationship between ciprofloxacin use and prevalence of ESBL Klebsiella pneumoniae]. / Alta correlación entre el consumo de ciprofloxacina y la prevalencia de Klebsiella pneumoniae productora de beta-lactamasas de espectro extendido.

BACKGROUND: To assess the relationship between ciprofloxacin use and the prevalence of extended spectrum betalactamases (ESBL) Klebsiella pneumoniae. PATIENTS AND METHODS: Semestral mean values regarding use of antibiotic and prevalence of ESBL Kp were compared during 9 semesters using linear regression and coefficient of correlation. RESULTS: The only statistically significant correlation was ciprofloxacin use and ESBL(+) K. pneumoniae prevalence, with a coefficient of correlation of 0.86 and p = 0.0027 using linear regression. CONCLUSIONS: Ciprofloxacin use must be taking into account when considering infection control programs due to high prevalence rates of ESBL(+) K. pneumoniae in the hospital setting.

Alta correlación entre el consumo de ciprofloxacina y la prevalencia de Klebsiella pneumoniae productora de ß-lactamasas de espectro extendido / High relationship between ciprofloxacin use and prevalence of ESBL Klebsiella pneumoniae

Background: To assess the relationship between ciprofloxacin use and the prevalence of extended spectrum betalactamases (ESBL) Klebsiella pneumoniae. Patients and Methods: Semestral mean values regarding use of antibiotic and prevalence of ESBL Kp were compared during 9 semesters using linear regression and coefficient of correlation. Results: The only statistically significant correlation was ciprofloxacin use and ESBL(+) K. pneumoniae prevalence, with a coefficient of correlation of 0.86 and p = 0.0027 using linear regression. Conclusions: Ciprofloxacin use must be taking into account when considering infection control programs due to high prevalence rates of ESBL(+) K. pneumoniae in the hospital setting.

Fundamento: Evaluar la correlación entre el consumo de cefalosporinas de tercera generación y ciprofloxacina con la prevalencia de cepas de Klebsiella pneumoniae productoras de ß-lactamasas de espectro extendido (BLEE). Pacientes y Métodos: Los valores promedios semestrales, correspondientes a consumo y prevalencia se compararon durante 9 semestres, usando coeficiente de correlación y regresión lineal. Resultados: La única asociación que resultó estadísticamente significativa, fue la correspondiente al consumo de ciprofloxacina y K. pneumoniae BLEE (+), con un coeficiente de correlación de 0,86 y una p de 0,0027, en el análisis de regresión lineal. Conclusiones: El consumo de ciprofloxacina debe ser tenido en cuenta al momento de establecer programas de control de infecciones frente a elevadas tasas de prevalencia de K. pneumoniae productoras de BLEE en un hospital.

[Risk factors associated with ceftazidime-resistant Klebsiella pneumoniae infection]. / Factores de riesgo asociados a infecciones por Klebsiella pneumoniae resistentes a ceftacidima.

INTRODUCTION: Risk factors associated with ceftazidime-resistant Klebsiella pneumoniae (CAZ-R Kp) infection may vary among hospitals and in the same hospital at different time points. Knowledge of these factors is required to establish suitable infection control programs. METHODS: A case-control study was conducted to assess risk factors for CAZ-R Kp infection. Thirty-two cases were compared with 28 controls admitted to a 200-bed general hospital during 1999 and 2000. RESULTS: In the univariate analysis Kp CAZ-R isolates were significantly associated with nosocomial acquisition (OR 5 17.40), prior antibiotic use (OR 5 14.94), particularly ciprofloxacin use (OR 5 5), and hospitalization stay of more than 6 days (OR 5 6.72). Significantly associated variables in the logistic regression analysis included nosocomial acquisition (OR 5 9.29), prior antibiotic use (OR 5 6.21), and particularly, ciprofloxacin use (OR 5 10.84). CONCLUSIONS: Efforts toward more rational overall antibiotic use and particularly ciprofloxacin use, combined with infection control measures are necessary to decrease the prevalence of CAZ-R Kp in our hospital.

Factores de riesgo asociados a infecciones por Klebsiella pneumoniae resistentes a ceftacidima / Risk factors associated with ceftazidime-resistant Klebsiella pneumoniae infection

INTRODUCCIÓN. Los factores de riesgo asociados a las infecciones debidas a Klebsiella pneumoniae resistentes a ceftacidima (CAZ-R) pueden variar entre las instituciones y en el tiempo. El conocimiento de factores propios son de relevancia para adecuar los programas de control de infecciones de los diferentes hospitales. MÉTODOS. Se utilizó un estudio de casos y controles para investigar los asociados a las infecciones debidas a K. pneumoniae CAZ-R. Se compararon 32 casos con 28 controles ingresados en un hospital general de 200 camas y complejidad intermedia, durante los años 1999 y 2000.RESULTADOS. En el análisis univariado el aislamiento de K. pneumoniae CAZ-R se asoció significativamente con la adquisición nosocomial (odds ratio [OR], 17,40), uso previo de antibióticos (OR, 14,94) especialmente ciprofloxacino (OR, 5) y el tiempo ingreso-muestra superior a 6 días (OR, 6,72). Al aplicar regresión logística sólo alcanzaron significación la adquisición nosocomial (OR, 9,29), el uso previo de antibióticos (OR, 6,21) y particularmente el uso de ciprofloxacino (OR, 10,84).CONCLUSIONES. Nuestro hospital debe realizar esfuerzos tendientes a reducir el consumo de antibióticos, sobre todo ciprofloxacino, junto con otras medidas de control de infecciones como estrategia para reducir la prevalencia de K. pneumoniae CAZ-R (AU)