Application and computer programs for a simple adaptive two dimensional smoother : a case study for cardiac procedure and death rates

Abstract

Age- and year- specific rates are widely used in epidemiological modelling studies. As these rates are usually unstable due to small denominators, these require smoothing in both dimensions. We demonstrated the application of a two dimensional nearest neighbour method for smoothing age- and year- specific cardiac procedure and death rates. SAS macros were provided for smoothing two rates successively, however these can be adapted to smooth more than two rates or event counts, if required. We found that for the example data sets, the order of the moving average in both year and age dimensions was three and hence a nine point weighted moving average was justified. We demonstrated that in terms of better calibration and capturing important changes in data, the proposed smoother outperformed a similar smoother assigning maximum weight to the central cell but equal weights around it. The degree of smoothing increased with increase in the assigned central cell weight. In conclusion, because of its simplicity, the proposed nearest neighbour smoother provides a convenient alternative to the existing two dimensional smoothers and is useful in situations requiring smoothing a series of rates or counts in two dimensions. A robust version of the smoother is also available from the author.