The effect of the look-back period for estimating incidence using administrative data

Abstract

Background: The look-back period is needed to define baseline population for estimating incidence. However, short look-back period is known to overestimate incidence of diseases misclassifying prevalent cases to incident cases. The purpose of this study is to evaluate the impact of the various length of look-back period on the observed incidences of uterine leiomyoma, endometriosis and adenomyosis, and to estimate true incidences considering the misclassification errors in the longitudinal administrative data in Korea. Methods: A total of 319,608 women between 15 to 54 years of age in 2002 were selected from Korea National Health Insurance Services (KNHIS) cohort database. In order to minimize misclassification bias incurred when applying various length of look-back period, we used 11 years of claim data to estimate the incidence by equally setting the look-back period to 11 years for each year using prediction model. The association between the year of diagnosis and the number of prevalent cases with the misclassification rates by each look-back period was investigated. Based on the findings, prediction models on the proportion of misclassified incident cases were developed using multiple linear regression. Results: The proportion of misclassified incident cases of uterine leiomyoma, endometriosis and adenomyosis were 32.8, 10.4 and 13.6% respectively for the one-year look-back period in 2003. These numbers decreased to 6.3% in uterine leiomyoma and − 0.8% in both endometriosis and adenomyosis using all available look-back periods (11 years) in 2013. Conclusion: This study demonstrates approaches for estimating incidences considering the different proportion of misclassified cases for various length of look-back period. Although the prediction model used for estimation showed strong R-squared values, follow-up studies are required for validation of the study results.

Evaluating the impact of covariate lookback times on performance of patient-level prediction models

“…Studies of lookback periods have been conducted in the context of incidence estimation/phenotype development, effect estimation, with a limited number of studies in the context of prediction [ 2 – 7 ]. Incidence estimation and effect estimation studies all advise that longer lookback reduces data noise for the diseases and data sources evaluated [ 2 – 4 ]. The findings from the relatively few prediction studies stated that for the outcomes evaluated that lookback period did not materially impact the model discrimination but did improve identification of the disease history [ 5 – 7 ].…”

“…Studies using administrative data and investigating variations in the length of lookback period have been conducted in the context of incidence and effect estimation [ 2 – 4 ]. In a study of cancer cumulative incidence estimation the authors recommended using lookback of 2 or more years and discouraged the use of 1 year lookback but caveated that it is not possible to provide general recommendations as lookback period is dependent on the characteristics of the cancer site and the available data and the underlying research question [ 3 ].…”

“…In a study of cancer cumulative incidence estimation the authors recommended using lookback of 2 or more years and discouraged the use of 1 year lookback but caveated that it is not possible to provide general recommendations as lookback period is dependent on the characteristics of the cancer site and the available data and the underlying research question [ 3 ]. A Korean study using a cohort database and examining lookback and estimating incidence of three gynecological diseases (uterine leiomyoma, endometriosis, and adenomyosis) found that as the lookback increased the proportion of misclassified incident cases decreased but advised that the optimal lookback for annual incidence depended on the nature and the stage of the respective diseases [ 4 ]. A comparative effect study using the Medicare beneficiary database and evaluating the effect of statin initiation on incidence of cancer recommended that a 3 year lookback was best but if infeasible that all available lookback is preferable to short fixed lookbacks [ 2 ].…”

Evaluating the Impact of Covariate Lookback Times on Performance of Patient-Level Prediction Models

“…[3]Studies using administrative data and investigating variations in the length of lookback period have been conducted in the context of incidence and effect estimation [4][5][6]. In a study of cancer cumulative incidence estimation the authors recommended using lookback of two or more years and discouraged the use of one year lookback but caveated that it is not possible to provide general recommendations as lookback period is dependent on the characteristics of the cancer site and the available data and the underlying research question [5].…”

“…In a study of cancer cumulative incidence estimation the authors recommended using lookback of two or more years and discouraged the use of one year lookback but caveated that it is not possible to provide general recommendations as lookback period is dependent on the characteristics of the cancer site and the available data and the underlying research question [5]. A Korean study using a cohort database and examining lookback and estimating incidence of three gynecological diseases (uterine leiomyoma, endometriosis, and adenomyosis) found that as the lookback increased the proportion of misclassi ed incident cases decreased but advised that the optimal lookback for annual incidence depended on the nature and the stage of the respective diseases [6]. A comparative effect study using the Medicare bene ciary database and evaluating the effect of statin initiation on incidence of cancer recommended that a three year lookback was best but if infeasible that all available lookback is preferable to short xed lookbacks [4].…”