A comparison of different methods for handling measurements affected by medication use

Abstract

In epidemiological research it is common to encounter measurements affected by medication use, such as blood pressure lowered by antihypertensive drugs. When one is interested in the relation between the variables not affected by medication, ignoring medication use can cause bias. Several methods have been proposed, but the problem is often ignored or handled with generic methods, such as excluding individuals on medication or adjusting for medication use in the analysis. This study aimed to investigate methods for handling measurements affected by medication use when one is interested in the relation between the unaffected variables and to provide guidance for how to optimally handle the problem. We focused on linear regression and distinguish between the situation where the affected measurement is an exposure, confounder or outcome. In the Netherlands Epidemiology of Obesity study and in several simulated settings, we compared generic and more advanced methods, such as substituting or adding a fixed value to the treated values, regression calibration, censored normal regression, Heckman’s treatment model and multiple imputation methods. We found that often-used methods such as adjusting for medication use could result in substantial bias and that methods for handling medication use should be chosen cautiously.

How measurements affected by medication use are reported and handled in observational research: A literature review

“…Additionally, no study was found to use any of the more advanced statistical methods previously suggested. 2,3,5,10 This may call for methodological papers in clinical journals that provide practical guidelines and tutorials on when and how to apply corrections for medication use in applied clinical research.…”

“…However, inadequately handling medication use in confounding variables can lead to bias. 10 We noticed that recommendations in methodological papers were seldom applied. For example, Tobin et al 2 recommended adding a constant value to measurements of treated values of an outcome variable when interest is in the underlying unaffected values and stressed the necessity of sensitivity analysis to determine the robustness to the particular choice of constant.…”

“…The assessment on whether the methods used are in general valid or invalid was based on recommendations from previous methodological studies. [2][3][4][5][6]10,11,[17][18][19][20][21][22][23][24][25][26] For example, restricting the study population to non-medication users was considered valid when the exposure or a confounder is affected by medication use regardless of whether the research aim is in the values as observed or if untreated. This is because the restriction on a proxy variable of the Exceed 100% when added up, because more than one analysis was performed in some studies.…”

“…Several methods for handling medication use have been proposed and compared. [1][2][3][4][5][6][7][8][9] Studies have shown that different methods may lead to substantially different effect estimates, [2][3][4][5][8][9][10] and the optimal method depends on (a) the research aim and (b) whether the medication effect is on the exposure, outcome or a confounder. 10 If the method used for handling medication effect does not match the research question, substantial bias can be introduced and the interpretation of results will be unclear.…”

“…[1][2][3][4][5][6][7][8][9] Studies have shown that different methods may lead to substantially different effect estimates, [2][3][4][5][8][9][10] and the optimal method depends on (a) the research aim and (b) whether the medication effect is on the exposure, outcome or a confounder. 10 If the method used for handling medication effect does not match the research question, substantial bias can be introduced and the interpretation of results will be unclear. 11 Thus, it is essential to carefully think about the research question when some individuals in a study population use medication that affects the variables in the dataset.…”