A common error in the ecological regression of cancer incidence on the deprivation index.

OBJECTIVE: To determine if introducing age as another explanatory variable in an ecological regression model relating crude rates of cancer incidence and a deprivation index provides better results than the usual practice of using the standard incidence ratio (SIR) as the response variable, introducing the non-standardized index, and not including age in the model. METHODS: Relative risks associated with the deprivation index for some locations of cancer in Spain's Girona Health Region were estimated using two different models. Model 1 estimated relative risks with the indirect method, using the SIR as the response variable. Model 2 estimated relative risks using age as an explanatory variable and crude cancer rates as the response variable. Two scenarios and two sub-scenarios were simulated to test the properties of the estimators and the goodness of fit of the two models. RESULTS: The results obtained from Model 2's estimates were slightly better (less biased) than those from Model 1. The results of the simulation showed that in all cases (two scenarios and two sub-scenarios) Model 2 had a better fit than Model 1. The probability density for the parameter of interest provided evidence that Model 1 leads to biased estimates. CONCLUSIONS: When attempting to explain the relative risk of incidence of cancer using ecological models that control geographic variability, introducing age as another explanatory variable and crude rates as a response variable provides less biased results.

A common error in the ecological regression of cancer incidence on the deprivation index / Un error frecuente en los modelos de regresión ecológica de la incidencia de cáncer con respecto al índice de carencia

OBJECTIVE: To determine if introducing age as another explanatory variable in an ecological regression model relating crude rates of cancer incidence and a deprivation index provides better results than the usual practice of using the standard incidence ratio (SIR) as the response variable, introducing the non-standardized index, and not including age in the model. METHODS: Relative risks associated with the deprivation index for some locations of cancer in Spain's Girona Health Region were estimated using two different models. Model 1 estimated relative risks with the indirect method, using the SIR as the response variable. Model 2 estimated relative risks using age as an explanatory variable and crude cancer rates as the response variable. Two scenarios and two sub-scenarios were simulated to test the properties of the estimators and the goodness of fit of the two models. RESULTS: The results obtained from Model 2's estimates were slightly better (less biased) than those from Model 1. The results of the simulation showed that in all cases (two scenarios and two sub-scenarios) Model 2 had a better fit than Model 1. The probability density for the parameter of interest provided evidence that Model 1 leads to biased estimates. CONCLUSIONS: When attempting to explain the relative risk of incidence of cancer using ecological models that control geographic variability, introducing age as another explanatory variable and crude rates as a response variable provides less biased results.

OBJETIVO: Determinar si la introducción de la edad como otra variable independiente en un modelo de regresión ecológica que relaciona las tasas brutas de incidencia de cáncer con un índice de carencia, ofrece mejores resultados que la práctica corriente del uso de la razón de incidencia normalizada como criterio de valoración, con introducción del índice sin normalización y sin incluir la edad en el modelo. MÉTODOS:Se calcularon los riesgos relativos asociados con el índice de carencia de algunos tipos de cáncer en la Región Sanitaria de Girona en España, mediante dos modelos diferentes. En el modelo 1 se calcularon los riesgos relativos con el método indirecto, usando la razón de incidencia normalizada como criterio de valoración. En el modelo 2 se calcularon los riesgos relativos introduciendo la edad como una variable independiente y las tasas brutas de cáncer como criterio de valoración. Se simularon dos hipótesis y dos subhipótesis con el fin de verificar las propiedades de los estimadores y la bondad del ajuste de ambos modelos. RESULTADOS: Los resultados obtenidos a partir de las estimaciones con el modelo 2 fueron un poco mejores (menos sesgados) que los resultados obtenidos con el modelo 1. Los resultados de la simulación indicaron que en todos los casos (las dos hipótesis y las dos subhipótesis) el modelo 2 exhibió un mejor ajuste que el modelo 1. La función de densidad del parámetro de interés puso en evidencia que el modelo 1 da lugar a estimaciones sesgadas. CONCLUSIONES: Cuando se intenta explicar el riesgo relativo de incidencia de cáncer mediante modelos de regresión ecológica que tienen en cuenta la variabilidad geográfica, se obtienen resultados menos sesgados cuando se introduce la edad como una de las variables independientes y se utilizan las tasas brutas de incidencia como criterio de valoración.

Desigualdades y mortalidad por cirrosis en varones (Zaragoza, 1996-2003) / Men inequalities and liver cirrhosis mortality (Zaragoza, Spain, 1996-2003)

Objetivos Analizar diferencias geográficas en la mortalidad por cirrosis hepática, en varones, en la provincia de Zaragoza, y su posible asociación con indicadores socioeconómicos, así como identificar la adecuación, en el ámbito rural, del índice de privación del proyecto MEDEA. Métodos La unidad geográfica de análisis para Zaragoza capital fue la sección censal, y para el resto de la provincia el municipio. Para cada unidad de análisis se calculó la razón de mortalidad estandarizada cruda y suavizada mediante un modelo lineal generalizado mixto bayesiano. Se calculó un índice sintético de privación y se incluyó en el modelo en cuartiles. También se realizó el análisis exploratorio incluyendo un índice de ruralidad para la provincia de Zaragoza.Resultados En Zaragoza capital, la mortalidad por cirrosis y otras enfermedades crónicas del hígado (código 571 de la 9ª revisión de la Clasificación Internacional de Enfermedades y códigos K70, K72.1, K73, K74, K76.1.9 de la 10ª revisión) se incrementaba a medida que aumentaba el índice de privación, y en las secciones censales más desfavorecidas era superior a la de las más favorecidas, con un riesgo relativo (RR) de 2,09 y un intervalo de credibilidad (IC) de 1,53-2,83. En el resto de la provincia, las diferencias en mortalidad no pueden explicarse por el índice de privación utilizado. En los municipios con valores más altos para el índice de ruralidad el RR fue de 0,47 (IC: 0,18-0,92) con respecto a aquellos que presentaron los valores más bajos.ConclusionesLas secciones censales del municipio de Zaragoza más deprimidas presentan una mayor mortalidad por cirrosis. Esta asociación no se ha encontrado en el resto de la provincia, posiblemente por la baja variabilidad explicada por el índice utilizado. Los municipios de la provincia con mayores valores del índice de ruralidad presentaron un menor riesgo de muerte por las causas en estudio (AU)

Objectives: The aim of this study was to identify geographical differences in mortality from liver cirrhosisin men living in the province of Zaragoza, Spain, as well as its possible association with socioeconomicfactors. The utility of the MEDEA project’s deprivation index in rural areas was also explored.Methods: Census tracts were used in Zaragoza city as analysis units and municipalities were used forthe rest of the province. Crude and smoothed standardized mortality ratios were calculated for eachanalysis unit through a Bayesian generalized mixed linear model. A deprivation index was obtained andwas included in the model in quartiles. An exploratory analysis was also conducted, including a ruralindex in the province of Zaragoza.Results: In Zaragoza city, mortality from liver cirrhosis and other chronic liver diseases [code 571 ofthe 9th International Classification of Diseases (ICD) and K70, K72.1, K73, K74, K76.1.9 of the ICD-10]increased as the deprivation index increased. Mortality in the most deprived areas was twice that in theless deprived areas (relative risk [RR] 2.09, credible interval (CI): 1.53-2.83). In the rest of the province,geographical differences in mortality could not be explained by the deprivation index used. Nevertheless,municipalities with the highest values in the rural index showed a RR of 0.47 (CI: 0.18-0.92) comparedwith those with the lowest values.Conclusions: In Zaragoza city, mortality from liver cirrhosis and other chronic liver diseases was higher inthe most deprived census tracts than in the most affluent areas. This association was not found in the restof the province, probably because of the low variability explained by the deprivation index. Municipalitieswith high rural values had the lowest risk of death from these diseases (AU)

[Men inequalities and liver cirrhosis mortality (Zaragoza, Spain, 1996-2003)]. / Desigualdades y mortalidad por cirrosis en varones (Zaragoza, 1996-2003).

OBJECTIVES: The aim of this study was to identify geographical differences in mortality from liver cirrhosis in men living in the province of Zaragoza, Spain, as well as its possible association with socioeconomic factors. The utility of the MEDEA project's deprivation index in rural areas was also explored. METHODS: Census tracts were used in Zaragoza city as analysis units and municipalities were used for the rest of the province. Crude and smoothed standardized mortality ratios were calculated for each analysis unit through a Bayesian generalized mixed linear model. A deprivation index was obtained and was included in the model in quartiles. An exploratory analysis was also conducted, including a rural index in the province of Zaragoza. RESULTS: In Zaragoza city, mortality from liver cirrhosis and other chronic liver diseases [code 571 of the 9th International Classification of Diseases (ICD) and K70, K72.1, K73, K74, K76.1.9 of the ICD-10] increased as the deprivation index increased. Mortality in the most deprived areas was twice that in the less deprived areas (relative risk [RR] 2.09, credible interval (CI): 1.53-2.83). In the rest of the province, geographical differences in mortality could not be explained by the deprivation index used. Nevertheless, municipalities with the highest values in the rural index showed a RR of 0.47 (CI: 0.18-0.92) compared with those with the lowest values. CONCLUSIONS: In Zaragoza city, mortality from liver cirrhosis and other chronic liver diseases was higher in the most deprived census tracts than in the most affluent areas. This association was not found in the rest of the province, probably because of the low variability explained by the deprivation index. Municipalities with high rural values had the lowest risk of death from these diseases.

Accurately estimating breast cancer survival in Spain: cross-matching local cancer registries with the National Death Index.

OBJECTIVES: To assess the impact of using data from the National Death Index (NDI) of Spain to estimate breast cancer survival rates among residents of Girona and Zaragoza diagnosed in 1995-1999. METHODS: This was an observational, longitudinal epidemiologic study, using two population-based cancer registries. Data collected were of female residents of Girona or Zaragoza who had been diagnosed with breast cancer in 1995-1999. Observed and relative 5-year survival rates were estimated, first using the information available from the Girona and Zaragoza cancer registries, and then with the inclusion of NDI data. The 5-year relative survival rate and corresponding 95% Confidence Intervals were estimated using the Hakulinen method. The Kaplan-Maier method and Log Rank test were used to compare survival curves. RESULTS: No statistically significant difference in survival curves was observed in Girona for the data obtained before and after cross-matching with the NDI. However, there was a significant difference in Zaragoza. A comparison of the relative survival rates of each of the two registries before NDI cross-matching showed differences of 3.9% (5-year) and 16.1% (10-year) between the two, whereas after the cross-match, the difference was only 0.5% (5-year) and 1.2% (10-year). CONCLUSIONS: In Spain it is imperative that there be systematic use of NDI data to supplement cancer registries, so that comparisons of relative survival rates between registries can be improved.

Accurately estimating breast cancer survival in Spain: cross-matching local cancer registries with the National Death Index / Estimación precisa de la supervivencia de cáncer de mama en España: emparejamiento de los registros locales de cáncer con el Índice Nacional de Defunciones

OBJECTIVES: To assess the impact of using data from the National Death Index (NDI) of Spain to estimate breast cancer survival rates among residents of Girona and Zaragoza diagnosed in 1995-1999. METHODS: This was an observational, longitudinal epidemiologic study, using two population- based cancer registries. Data collected were of female residents of Girona or Zaragoza who had been diagnosed with breast cancer in 1995-1999. Observed and relative 5-year survival rates were estimated, first using the information available from the Girona and Zaragoza cancer registries, and then with the inclusion of NDI data. The 5-year relative survival rate and corresponding 95 percent Confidence Intervals were estimated using the Hakulinen method. The Kaplan-Maier method and Log Rank test were used to compare survival curves. RESULTS: No statistically significant difference in survival curves was observed in Girona for the data obtained before and after cross-matching with the NDI. However, there was a significant difference in Zaragoza. A comparison of the relative survival rates of each of the two registries before NDI cross-matching showed differences of 3.9 percent (5-year) and 16.1 percent (10-year) between the two, whereas after the cross-match, the difference was only 0.5 percent (5-year) and 1.2 percent (10-year). CONCLUSIONS: In Spain it is imperative that there be systematic use of NDI data to supplement cancer registries, so that comparisons of relative survival rates between registries can be improved.

OBJETIVO: Evaluar el efecto de utilizar los datos del Índice Nacional de Defunciones (IND) de España para estimar las tasas de supervivencia de cáncer de mama en las mujeres residentes en Girona y Zaragoza que recibieron el diagnóstico de cáncer de mama en 1995-1999. MÉTODOS: Se realizó un estudio epidemiológico observacional y longitudinal basado en el empleo de los registros de cáncer de mujeres residentes en Girona y Zaragoza que habían recibido el diagnóstico de cáncer de mama en 1995-1999. Se estimaron las tasas de supervivencias observada y relativa a 5 años, primero según la información disponible en los registros de cáncer de Girona y Zaragoza y luego con la inclusión de los datos del IND. Se calcularon las tasas de supervivencia relativa a 5 años y sus correspondientes intervalos de confianza de 95 por ciento por el método de Hakulinen. Las curvas de supervivencia se compararon por el método de Kaplan-Maier y la prueba de rangos logarítmicos. RESULTADOS: No se encontraron diferencias estadísticamente significativas entre las curvas de supervivencia de Girona antes y después de emparejar lo datos locales con los del IND; sin embargo, hubo diferencias significativas entre las curvas de Zaragoza. Al comparar las tasas de supervivencia relativa de cada uno de los registros antes de emparejarlos con los datos del IND se encontraron diferencias de 3,9 por ciento (a 5 años) y 16,1 por ciento (a 10 años), mientras que después del emparejamiento, la diferencia entre ellas fue solamente de 0,5 por ciento (a 5 años) y 1,2 por ciento (a 10 años). CONCLUSIONES: En España es imperativo el empleo sistemático de los datos del IND para complementar los registros de cáncer de manera de mejorar las comparaciones de las tasas de supervivencia relativa cuando se utilizan diferentes registros.