What Opportunities AI Offers for Clinical Research
Published: 06.02.2021
Artificial intelligence (AI) and machine learning have become increasingly important in business, science and society in recent years. AI also offers a wide range of applications and opportunities for clinical research. An overview.
AI - what is artificial intelligence?
Generally speaking, artificial intelligence (AI) is used when a device or algorithm uses its environment. Either decisions are made or actions are executed in order to maximise the chances of achieving a defined goal. Use cases for this can be in the area of recognition or planning, for example.
The problem with this definition of artificial intelligence: As soon as a complex problem has been solved, a new milestone in this field has been reached, the programme or algorithm is no longer perceived as AI. An example of this is traffic sign recognition in modern cars. Although this is a complex task, no real intelligence is needed to check whether images look identical or not. This is the so-called AI effect, which leads to another definition of artificial intelligence: AI is what no one has managed to do yet.
These two definition approaches do not cover the entire research area of AI. More precise definitions exist for some areas. Nevertheless, this shows the omnipresent dichotomy of the question of whether a new programme or algorithm is AI or not.
In the field of clinical research, the goal of artificial intelligence is self-expanding, self-learning software for complex pattern recognition and evaluation.
Latest Whitepaper
What does AI have to do with clinical research?
Clinical research is based on data that document individual therapeutic procedures and observation processes. Therefore, usable results require clean, complete data and good data management. This is a complex, time-consuming process that continuously accompanies data documentation within a clinical trial.
For example, the analysis of survival data, the search for correlations between different characteristics, therapy successes and failures as well as concomitant medications and side effects are mostly carried out only at the end of a study on the basis of a previously defined analysis plan. Although various reports or interim evaluations are collected between the start and end of a clinical trial, these are of a much smaller scope than the final evaluation.
Good specifications and data checking plans are an enormous help towards the end of data collection, but at the same time this period remains a potential source of errors. This is when a well-designed AI provides valuable support by pre-emptively preventing a large part of the questions and inaccuracies that arise.
The advantage of AI: A data management system supported by this tool reacts much more agilely and reliably, resulting in maximally clean and reliable data in optimal study conditions.
Artificial intelligence: First steps in development
The conception and development of good AI requires preparatory work. Users should ask themselves the following questions, among others:
-
How detailed is the information that the AI is supposed to work with usually available?
-
How reliable is the data?
-
How accurate can information be when it is not measured values but memories of an individual?
-
Do different pieces of information need to be weighted differently?
-
How are the data related and can certain information be extracted more reliably elsewhere?
Only by analysing all the data as closely as possible and answering many specific questions are developers able to identify the most important features of an AI product for continuous optimisation.
Even though the preparatory data analysis is one of the first steps in the development process of an artificial intelligence based on statistical methods, the first insights are already being gained. When documenting categorical variables, such as a medication type, certain options are displayed for selection in an eCRF (electronic case report form). If none of the given options fits, the entry "Other" usually follows. This usually results in a free text field in which more detailed information is entered.
A system that is prone to errors: In the example of medication, the corresponding entry in the selection could be overlooked, "Other" selected and the medication entered as free text. Possibly this is done with a spelling error.
To avoid such impurities in the data, an automated, minimalist search algorithm is interposed, which matches entered free texts with the selection options, even taking typing errors into account. If a hit is found, a corresponding warning is displayed.
This simple example shows how a relief of the data management and a higher degree of purity of the documented data is achieved in the first instance.