Methods for using clinical laboratory test results as baseline confounders in multi‐site observational database studies when missing data are expected

Abstract

Multi-site studies must consider site variability in missing data. Different missing data methods performed similarly. Copyright © 2016 John Wiley & Sons, Ltd.

Characteristics of Hospital Differences in Missing of Clinical Laboratory Test Results in a Multi-hospital Observational Database Contributing to MID-NET® in Japan

“…The first and second periods were adopted by referring to previous cohort studies using a laboratory test as baseline covariate [13][14][15] and a previous study assessing missing data in the MSDD for 183 days. 9 The third and fourth periods were adopted for cases where all outcomes were included and for cases where the study interest was the only outcome after a certain period from the prescription date, respectively. The 365 days in the fourth period was created by referring to the mean follow-up period in a previous study of our scenario.…”

“…Seven cohort study scenarios using laboratory results as baseline covariates were created (Supplementary Figure S1). Scenarios 1-5 were original scenarios; scenarios 6 and 7 were incorporated to compare our results with those of Raebel et al 9 Scenario setting details are provided in Table 1. The backgrounds of the original scenarios 1-5…”

“…6,7 The appropriate use of these data is difficult because some data obtained during routine medical care may be missing in datasets for analysis. 8,9 Missing covariate data is a critical issue for observational studies requiring confounding adjustments. Various methods have been proposed to overcome improper handling of missing data that can result in bias.…”

“…8,10 The features of missing data (e.g., missing proportion and factors associated with missingness) and sources of missing data are crucial for choosing appropriate missing data methods. 9,10 Missing proportions and factors associated with missingness can differ across data partners in databases covering multiple sites or hospitals. The variability in missing data among data partners is a critical issue for applying the missing data method.…”

“…The variability in missing data among data partners is a critical issue for applying the missing data method. For three sites contributing to the US Food and Drug Administration Mini-Sentinel Distributed Database (MSDD), Raebel et al 9 reported that the missing proportion of baseline laboratory results and factors associated with missingness varied by site. Differential missingness across sites was attributed to multiple factors, such as the type of data partner (e.g., only with laboratory results of outpatients) and patient background.…”