Clinical Data Analysis (CDA)

The functional block of CDA facilitates the detailed study of data related to the physical condition and health of people in order to acquire or improve their knowledge. To this end, existing technologies and analysis techniques will be applied (DS, ML and Al, according to their definition and state of the art), with the following means:

■ Data: The clinical data will be obtained through downloads, ETL processes or other mechanisms of integration with the analysis tools, from the ODM, the EMR-OEMR and external data sources. These data will be pseudonymized in the process of obtaining so that the confidentiality of the information is protected to a greater extent and the possibility of participation of researchers not directly involved in patient care is enabled.

■ Methodology: Although the usual approach from the point of view of clinical research is to propose a hypothesis and then obtain the data to validate it, we observed how in our approach we had assistance data in the first place, and we considered how to obtain an additional value of them. Therefore, we started from the need to perform an analysis of clinical oncological data without specifying

a previous hypothesis or a series of specific objectives, but multiple areas of interest to explore through a methodology that allowed us to perform this process in an optimal and systematic way. Therefore, to carry out the data analyses necessary to achieve the proposed objective, CRISP-DM has been used as a methodological framework (Chapman et al., 2000). We have applied this methodology adapting it to the needs of the healthcare environment and the problem of analysis, taking as reference previous studies that apply this methodology to the analysis of clinical and healthcare data (Rivo et al., 2012) (Pérez et al., 2015). Thus, we propose the adoption of CRISP-DM with phases adapted and developed in an iterative life cycle until the desired results are obtained. We propose an inhomogeneous distribution of the data analysis project efforts in the different iterations of the CRISP-DM cycle. Thus, the first iterations will be more focused on obtaining clear analysis objectives and understanding them properly to obtain the necessary data from the available sources. The following iterations will focus more on the preparation of the data and the development of initial models for evaluation by clinical staff. Finally, the last iterations will try to focus their efforts on refining the models and evaluating them in detail, as well as obtaining results that can be exploited in the clinical setting (Figure 7.6).

■ Tools: Our proposal allows complete freedom when selecting the most appropriate technological tools to achieve the proposed objectives.

Figure 7.6 Succession of CRISP-DM cycles over time with an inhomogeneous distribution of efforts in their different phases. (Adapted from https://es.wikipedia.org/wiki/ Cross_lndustry_Standard_Process_for_Data_Mining.)

Thus, the final product of CDA will be the following:

■ Knowledge: Regarding the analysed data (a piece of evidence) that in many cases could be the starting point for new research hypotheses or for supporting clinical decision-making.

■ Mathematical models: Trained on the available data that could be deployed, after a proper validation, in the clinical environment (production) to support clinical decision-making.