Optimising cohort data in Europe

through data processing procedures. The main technical challenges found in this harmonisation strategy are the data quality issues that may be found across different studies, such as data incompleteness, potential misclassification of data, or ambiguous terminologies (Fortier et al., 2017). The efforts of this strategy are focused on being able to standardise the data of each cohort study retrospectively and to create bridges to harmonisation of the variables with minimum possible bias. The main challenges to the ex post retrospective approach are (Dubrow and Tomescu-Dubrow, 2016; Fortier et al., 2017): y y To find a balance between precision and number of harmonised variables across studies, in order to ensure correct inferences. y y To find ways to measure the quality of harmonisation. y y To define the kind of documentation of the involved studies required to ensure maximum efficiency in the harmonisation process. y y To disseminate the harmonisation process so that it is reproducible. y y To overcome a minimum investment in resources in terms of time and money, which is usually sought. 2.2. Analytical methods Harmonisation reduces the heterogeneity of comparable data across studies. We refer to reducing because heterogeneities can hardly be eliminated even in prospective harmonisation strategies. Therefore, harmonised data must be analysed to obtain integrated results, considering possible heterogeneities to correctly interpret differences in assessing an exposure or intervention effect of interest (Burke et al., 2017). Common heterogeneity sources, amongst others, are: y y Measurement characteristics such as levels of precision of each study. y y Different covariates collected. y y Differences between population samples. Therefore, proper modelling of the data should control for the specificities of the studies. Studies can be combined at the level of individual participant data (one-stage or pooled analysis) or at the level of their results or inferences (two-stage or meta-analysis). To conduct a pooled analysis, data from the different studies must necessarily meet two main requirements: the data must be available to the same researchers across the different studies, and they can be properly harmonised. The requirement that data can be harmonised depends very much on the homogeneity of the same data across studies, with the main challenge of inferences being as correct as possible. To do this, a y y Unmeasured confounders that vary by centre. y y Stage of the trials or conditions of the cohorts.