Effects of the choice of age‐adjustment method on maps of death rates

Linda Williams Pickle, Andrew A. White

Research output: Contribution to journalArticle

45 Citations (Scopus)

Abstract

Maps of morbidity or mortality rates, whether considered individually or as a layer in a geographic information system application, invite multiple comparisons of area rates. However, comparisons of rates across different populations require standardization of the age‐specific rates to account for differences in population age structures. The indirect standardization method, or equivalently the standardized mortality ratio (SMR), has been recommended for small areas where age‐specific rates can be quite variable. Although theoretically equivalent to directly adjusted rates under the assumption of independent age and area effects, indirect summary measures are not comparable across areas when this assumption is violated. We tested the validity of this assumption for the 10 most common causes of death in the United States during 1980–84 and examined the geographic clustering apparent when categorized death rates, adjusted by different methods, are presented as thematic maps. Although overall agreement between the methods was good (rank correlation coefficient >82 per cent for each cause), when the adjusted rates were classified into quintiles 18 per cent of the states fell into different categories depending on the method of adjustment. Using an internal standard for the indirect method reduced this discrepancy to 4·9 per cent. However, both traditional chi‐square tests and a generalized logistic spline model identified significant interactions between age and area for each cause of death, a violation of the assumption required for equivalence of the methods. Potential variation in geographic inferences is illustrated by maps of direct and indirect rates and an empirical Bayes posterior mean, which is a function of these traditionally adjusted rates. Based on these results, we recommend the direct age‐adjustment method for rate maps.

Original languageEnglish (US)
Pages (from-to)615-627
Number of pages13
JournalStatistics in Medicine
Volume14
Issue number5-7
DOIs
StatePublished - Jan 1 1995

Fingerprint

Mortality
Standardization
Quintile
Cause of Death
Posterior Mean
Age Structure
Spearman's coefficient
Population Structure
Chi-squared test
Multiple Comparisons
Empirical Bayes
Morbidity
Mortality Rate
Geographic Information Systems
Direct Method
Correlation coefficient
Logistics
Spline
Discrepancy
Adjustment

All Science Journal Classification (ASJC) codes

  • Epidemiology
  • Statistics and Probability

Cite this

Pickle, Linda Williams ; White, Andrew A. / Effects of the choice of age‐adjustment method on maps of death rates. In: Statistics in Medicine. 1995 ; Vol. 14, No. 5-7. pp. 615-627.
@article{fd6bcadb22304da6b88882d25b9ccfa8,
title = "Effects of the choice of age‐adjustment method on maps of death rates",
abstract = "Maps of morbidity or mortality rates, whether considered individually or as a layer in a geographic information system application, invite multiple comparisons of area rates. However, comparisons of rates across different populations require standardization of the age‐specific rates to account for differences in population age structures. The indirect standardization method, or equivalently the standardized mortality ratio (SMR), has been recommended for small areas where age‐specific rates can be quite variable. Although theoretically equivalent to directly adjusted rates under the assumption of independent age and area effects, indirect summary measures are not comparable across areas when this assumption is violated. We tested the validity of this assumption for the 10 most common causes of death in the United States during 1980–84 and examined the geographic clustering apparent when categorized death rates, adjusted by different methods, are presented as thematic maps. Although overall agreement between the methods was good (rank correlation coefficient >82 per cent for each cause), when the adjusted rates were classified into quintiles 18 per cent of the states fell into different categories depending on the method of adjustment. Using an internal standard for the indirect method reduced this discrepancy to 4·9 per cent. However, both traditional chi‐square tests and a generalized logistic spline model identified significant interactions between age and area for each cause of death, a violation of the assumption required for equivalence of the methods. Potential variation in geographic inferences is illustrated by maps of direct and indirect rates and an empirical Bayes posterior mean, which is a function of these traditionally adjusted rates. Based on these results, we recommend the direct age‐adjustment method for rate maps.",
author = "Pickle, {Linda Williams} and White, {Andrew A.}",
year = "1995",
month = "1",
day = "1",
doi = "10.1002/sim.4780140519",
language = "English (US)",
volume = "14",
pages = "615--627",
journal = "Statistics in Medicine",
issn = "0277-6715",
publisher = "John Wiley and Sons Ltd",
number = "5-7",

}

Effects of the choice of age‐adjustment method on maps of death rates. / Pickle, Linda Williams; White, Andrew A.

In: Statistics in Medicine, Vol. 14, No. 5-7, 01.01.1995, p. 615-627.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Effects of the choice of age‐adjustment method on maps of death rates

AU - Pickle, Linda Williams

AU - White, Andrew A.

PY - 1995/1/1

Y1 - 1995/1/1

N2 - Maps of morbidity or mortality rates, whether considered individually or as a layer in a geographic information system application, invite multiple comparisons of area rates. However, comparisons of rates across different populations require standardization of the age‐specific rates to account for differences in population age structures. The indirect standardization method, or equivalently the standardized mortality ratio (SMR), has been recommended for small areas where age‐specific rates can be quite variable. Although theoretically equivalent to directly adjusted rates under the assumption of independent age and area effects, indirect summary measures are not comparable across areas when this assumption is violated. We tested the validity of this assumption for the 10 most common causes of death in the United States during 1980–84 and examined the geographic clustering apparent when categorized death rates, adjusted by different methods, are presented as thematic maps. Although overall agreement between the methods was good (rank correlation coefficient >82 per cent for each cause), when the adjusted rates were classified into quintiles 18 per cent of the states fell into different categories depending on the method of adjustment. Using an internal standard for the indirect method reduced this discrepancy to 4·9 per cent. However, both traditional chi‐square tests and a generalized logistic spline model identified significant interactions between age and area for each cause of death, a violation of the assumption required for equivalence of the methods. Potential variation in geographic inferences is illustrated by maps of direct and indirect rates and an empirical Bayes posterior mean, which is a function of these traditionally adjusted rates. Based on these results, we recommend the direct age‐adjustment method for rate maps.

AB - Maps of morbidity or mortality rates, whether considered individually or as a layer in a geographic information system application, invite multiple comparisons of area rates. However, comparisons of rates across different populations require standardization of the age‐specific rates to account for differences in population age structures. The indirect standardization method, or equivalently the standardized mortality ratio (SMR), has been recommended for small areas where age‐specific rates can be quite variable. Although theoretically equivalent to directly adjusted rates under the assumption of independent age and area effects, indirect summary measures are not comparable across areas when this assumption is violated. We tested the validity of this assumption for the 10 most common causes of death in the United States during 1980–84 and examined the geographic clustering apparent when categorized death rates, adjusted by different methods, are presented as thematic maps. Although overall agreement between the methods was good (rank correlation coefficient >82 per cent for each cause), when the adjusted rates were classified into quintiles 18 per cent of the states fell into different categories depending on the method of adjustment. Using an internal standard for the indirect method reduced this discrepancy to 4·9 per cent. However, both traditional chi‐square tests and a generalized logistic spline model identified significant interactions between age and area for each cause of death, a violation of the assumption required for equivalence of the methods. Potential variation in geographic inferences is illustrated by maps of direct and indirect rates and an empirical Bayes posterior mean, which is a function of these traditionally adjusted rates. Based on these results, we recommend the direct age‐adjustment method for rate maps.

UR - http://www.scopus.com/inward/record.url?scp=0028986445&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0028986445&partnerID=8YFLogxK

U2 - 10.1002/sim.4780140519

DO - 10.1002/sim.4780140519

M3 - Article

VL - 14

SP - 615

EP - 627

JO - Statistics in Medicine

JF - Statistics in Medicine

SN - 0277-6715

IS - 5-7

ER -