Lau C Thygesen1, Cristobal Baixauli-Pérez2, Julián Librero-López2, Natalia Martínez-Lizaga3, Manuel Ridao-López3, Enrique Bernal-Delgado4. 1. 1 National Institute of Public Health, University of Southern Denmark, Copenhagen, Denmark. 2. 2 Fundación para el Fomento de la Investigación Sanitaria y Biomédica de la Comunidad Valenciana (FISABIO), Valencia, Spain 3 Red de Investigación en Servicios de Salud en Enfermedades Crónicas (REDISSEC), Spain. 3. 3 Red de Investigación en Servicios de Salud en Enfermedades Crónicas (REDISSEC), Spain 4 Institute for Health Sciences in Aragon (IIS Aragon), Zaragoza, Spain. 4. 3 Red de Investigación en Servicios de Salud en Enfermedades Crónicas (REDISSEC), Spain 4 Institute for Health Sciences in Aragon (IIS Aragon), Zaragoza, Spain ebernal.iacs@aragon.es.
Abstract
BACKGROUND: In geographical studies, population distribution is a key issue. An unequal distribution across units of analysis might entail extra-variation and produce misleading conclusions on healthcare performance variations. This article aims at assessing the impact of building more homogeneous units of analysis in the estimation of systematic variation in three countries. METHODS: Hospital discharges for six conditions (congestive heart failure, short-term complications of diabetes, hip fracture, knee replacement, prostatectomy in prostate cancer and percutaneous coronary intervention) produced in Denmark, England and Portugal in 2008 and 2009 were allocated to both original geographical units and new ad hoc areas. New areas were built using Ward's minimum variance methods. The impact of the new areas on variability was assessed using Kernel distribution curves and different statistic of variation such as Extremal Quotient, Interquartile Interval ratio, Systematic Component of Variation and Empirical Bayes statistic. RESULTS: Ward's method reduced the number of areas, allowing a more homogeneous population distribution, yet 20% of the areas in Portugal exhibited less than 100 000 inhabitants vs. 7% in Denmark and 5% in England. Point estimates for Extremal Quotient and Interquartile Interval Ratio were lower in the three countries, particularly in less prevalent conditions. In turn, the Systematic Component of Variation and Empirical Bayes statistic were slightly lower in more prevalent conditions. CONCLUSIONS: Building new geographical areas produced a reduction of the variation in hospitalization rates in several prevalent conditions mitigating random noise, particularly in the smallest areas and allowing a sounder interpretation of the variation across countries.
BACKGROUND: In geographical studies, population distribution is a key issue. An unequal distribution across units of analysis might entail extra-variation and produce misleading conclusions on healthcare performance variations. This article aims at assessing the impact of building more homogeneous units of analysis in the estimation of systematic variation in three countries. METHODS: Hospital discharges for six conditions (congestive heart failure, short-term complications of diabetes, hip fracture, knee replacement, prostatectomy in prostate cancer and percutaneous coronary intervention) produced in Denmark, England and Portugal in 2008 and 2009 were allocated to both original geographical units and new ad hoc areas. New areas were built using Ward's minimum variance methods. The impact of the new areas on variability was assessed using Kernel distribution curves and different statistic of variation such as Extremal Quotient, Interquartile Interval ratio, Systematic Component of Variation and Empirical Bayes statistic. RESULTS: Ward's method reduced the number of areas, allowing a more homogeneous population distribution, yet 20% of the areas in Portugal exhibited less than 100 000 inhabitants vs. 7% in Denmark and 5% in England. Point estimates for Extremal Quotient and Interquartile Interval Ratio were lower in the three countries, particularly in less prevalent conditions. In turn, the Systematic Component of Variation and Empirical Bayes statistic were slightly lower in more prevalent conditions. CONCLUSIONS: Building new geographical areas produced a reduction of the variation in hospitalization rates in several prevalent conditions mitigating random noise, particularly in the smallest areas and allowing a sounder interpretation of the variation across countries.