From “Big Data” to the exploration of associations in the space of morbidity and care

Of all the discussions about “Big Data”, we are only interested in a small excerpt here: the ideas that have been made explicit on this topic in German health research or that have implicitly influenced the discussion. It is important to examine this section in more detail because biomedical research worldwide is in the process of significantly expanding its range of instruments. Unfortunately, the term “big data” does little to help us understand this process – on the contrary. It is about the historically unprecedented opportunity to observe large numbers of people with regard to the natural development of their health and to study the application and impact of interventions available on a mass scale. This approach has potential that has not yet been realized. The question of what access Germany will find to this development arises precisely from this perspective.

>> "Big Data" is not a term, certainly not a scientific one. The combination of the two words is a buzzword, a catchphrase, a slogan. It was brought into the world - how could it be otherwise - by an American management consultancy. Gartner Inc. tweeted the following 2011: “Big Data is high-volume, high-velocity and/or high-variety information assets that demand cost-effective, innovative forms of information processing that enable enhanced insight, decision making, and process automation.” The impetus for this came from e-commerce and is to be understood in this context: "In 2001/02, IT organizations must look beyond traditional direct brute-force physical approaches to data management. Through 2003/04, practices for resolving e-commerce accelerated data volume, velocity, and variety issues will become more formalized/diverse. Increasingly, these techniques involve tradeoffs and architectural solutions that involve/impact application portfolios and business strategy decisions.” How can companies learn more about their customers and their own processes from the data that develops from their e-commerce activities Benefit for yourself? This is what “Big Data”1 is all about.
We shouldn't actually use the term in health research.
The “3 V's”2, which are quoted in many German publications today, come from this context. Companies should deal with three phenomena if they want to get more out of their data: Large amounts have to be analyzed, analyzes are needed more quickly than was previously the case, and the restriction to the narrow cost of SQL-tamed data should be overcome: The statistical programs could To date, we have only processed a limited amount of data sets - but there were increasingly many more. The analyzes used to take an eternity with the available resources - but it had to be done much faster (e.g. when a specific customer appeared on the Internet) and there was increasing information that did not fit the well-known architecture of SQL-based programs - it It was discovered that images and texts contained far too much unused information to be left unused. Photos, for example, which can now be automatically classified with regard to important properties (see example below). You have to master “volume”, “velocity” and “variety” if you want to keep up with Amazon in e-commerce. "Big Data" is the slogan that was created to make the problem clear to consulting clients: to quickly analyze large numbers of cases with different types of data - hardware, software and the organization of knowledge must be adapted to this. “Big Data” has become an extremely successful slogan to this day. There is nothing about this that should upset a German health researcher.
Reception of “Big Data” in German
Health research
But still does it to this day. The excitement is actually in the air because there is almost nothing comparable to e-commerce in the German healthcare system. There are no health-related platforms where various data sources from millions of people would be available. There are no business models where data could be subjected to such procedures in order to make faster and more efficient decisions. The excitement is sparked by the fiction that this data exists and that it could be used to steer the healthcare system in much the same way that Amazon uses its knowledge to make customers better customers. If that were the case, one would have to deal with the following challenges:
1. Abuse through manipulation and discrimination against platform users.
2. Unfair distribution of income from the Internet economy
3. Waste of resources because the analyzes mostly show nonsense.
4. Unfair monopolization of information by Internet companies.
5. Violation of common rules of scientific thought.

The first three allegations relate more or less to ethical aspects. They criticize a business model that does not exist in the healthcare sector in Germany (at least not yet). This complex will not be discussed further below, although it is clear that the ethical criticism of successful companies in the Internet economy is essentially political, which contributes to a considerable extent to the overall negative attitude towards the phenomenon of “big data”. The fourth accusation, on the other hand, expresses something like a secret longing for the analysis of the data mentioned and is therefore also an element of the general ambivalence towards this topic3.
Rather, our attention is focused on the last-mentioned accusation that dealing with large amounts of data could undermine scientific rigor or even science. This point of criticism is of far greater significance because – as will be shown later – it also affects research that would already be possible today.
What it's about: Chris Anderson, a journalist from the USA, published an article in a popular IT journal in 2008, which was particularly popular in Germany under the title “The End of Theory: The Data Deluge Makes the Scientific Method Obsolete”. cited article. The piece blatantly attacks the position of the scientific mainstream and disrespectfully postulates that in the future one will only have to read from the data and no longer have to struggle to use theory to unlock their secrets. Anderson quotes the then head of research at Google: “All models are wrong, and increasingly you can succeed without them” and comes to the conclusion “Google was right”.
These half-strong tones from the Internet economy are particularly interesting because they could cause shivers in the health sciences scene in Germany: “The possibilities of the new world announced under the heading of Big Data almost completely disregard the usual cornerstones of science. It is announced that the age of causality is over and that now only correlations are important. There is no need to build theories, the raw material and the method are sufficient data.”4 [2]
With the proclamation of the "Petabyte Age", in which the laws of scientific gravity should no longer apply, Chris Anderson has risen to the epistemological Lucifer. Why an American specialist journalist can make German health scientists ponder so much is perhaps understandable by the fact that large Internet companies are believed to be quite capable of being right in their assumption that they can only change the world with data and without any theory. And perhaps the proclamation of “Correlation is enough” [1] has shaken the prevailing scientific paradigm a little.
Otherwise, it must be stated that “Big Data” is given the status of an independent method, but it is not specified what this method is actually supposed to be. Essentially, the various attempts at definition are based on the fact that data should be analyzed that was previously not considered suitable for health science analyses: data from mass internet-based business processes. Analyzing this data, however, means nothing more than looking for conspicuous features of their joint occurrence, for patterns, chains and the like. This is actually the core of “Big Data”. All V's only describe that new analysis tools are being developed that can process more and different data in a shorter time5.
The core of “Big Data”: transactional data
So is there nothing special about “big data” other than “big”? Yes, but it's not the speed or the variety. What makes it special is that the source of the information is not surveys, public statistics, censuses or other known sources, but that it is derived directly from the various transactions6 on the Internet - business, social, cultural. What these now countless sources have in common is that they leave traces of data that can provide information about the motives of users in general and in individual cases, without this data having been explicitly generated for the purpose of the question. This is followed by a universe of questions with an infinite number of social, economic and health science facets.
The book “Dataclysm” therefore also has the apt English subtitle “Who We Are When We Think No One's Looking.” [13] It shows the diversity and significance of such analyzes using the example of data from an internet-based dating agency7. Two examples show this:
1. By the age of 40, women are most attracted to men around their own age. After the age of 40, the most attractive age is slightly below. In contrast, men between the ages of 20 and 50 generally find 20-year-old women most attractive.
2. White women and white men find potential black partners significantly less attractive than average.
The analyzes reveal behaviors and attitudes that respondents would normally never commit to because it is not politically correct to have such attitudes. What man would admit that he secretly only likes 20-year-old women? What woman would admit that she finds black men far less attractive than white men?8 Who would deny that such insights are useful for studying racist attitudes and fill a gap in our knowledge?
The examples quoted are a tiny excerpt from the possibilities for analyzing transaction data from the Internet. All major Internet companies now also offer such data for scientific use or have their own units10. "Like Google, a gaggle of social media networks such as Facebook, Twitter, LinkedIn, Foursquare, and others sit on an enormous treasure chest of datafied information that, once analyzed, will shed light on social dynamics at all levels, from the individual to society at large.” [10] Nevertheless, it must be assumed that only a fraction of the information sees the light of day. Most of it is likely to be used for private-sector operational purposes.
What the examples also show is the fact that the analyzes were not carried out in the classic ways of documented social scientific studies. Nevertheless, one should not give in to the idea that everything is correlated with everything else and that every meaningless connection is spread11. Of course, analysts from the Internet industry also know the scientifically devastating potential of confounding and develop sophisticated methods to curb it. [13]
Why “Correlation is enough”
tempers so heated
The examples also show that the search for connections between phenomena is at the core of the use of transaction data from the Internet – i.e. “big data”. “Correlation is enough” is the exaggerated description of this. “Correlation-based exploration” would be a more appropriate name for this activity. Or: “The search for abnormalities.” At this point, however, the approach comes into conflict with the basic scientific understanding of health science.
The prevailing paradigm in health research can be read in detail in the numerous memoranda on health services research. These were initiated by the "Network Health Services Research", which organizes practically the entire academic health science. Although one cannot do justice to the depth of the scientific positions represented in the memoranda with a few sentences, it can be read that “health science” can agree on the epistemological position of “critical rationalism”. As is well known, this had formed in the confrontation with positivism and in this very old debate sided with the deductionists. In contrast to inductivism, the datum that can be experienced with the senses means practically nothing in critical rationalism, because it cannot be thought of without using concepts that already contain theoretical assumptions about the functioning of the world. [12] In “Memorandum IV” it reads like this: “Theory is the prerequisite for the scientific perspective of empirical research. Empirical data only gain their relevance through theory.” [4]
The existing epistemological constitution of the health sciences has its greatest impact on the question of how data should be handled. Can you just take them as they are and analyze them without first asking the exact question of what you are actually looking for and when can you say that you have answered the question posed? "No", say the health and health services researchers: "In all studies on health services research, target variables, end points or decision criteria must be defined, justified, comprehensibly documented and presented transparently as far as possible a priori." [11]12
It is clear that in this scientific world view, correlation as the core of the search for connection is at best of secondary importance. In the criticism of "Big Data" it is suggested that analyzes based on correlations are not serious analyses, that the scientific claim to causality would be challenged by the correlation.[2]
This concern can also be found in relation to health services research: “… the big data discussion is based on its historical context; Otherwise, health services research will degenerate into mere data accumulation.” [14] It fits with this distant position that health services research has predominantly taken a warning, but not offensive, approach to transactional data research. This is also made clear in Memorandum IV: “In view of the information technology revolution, which will affect medicine not only in terms of the explosion in data volumes, the topic of 'Big Data' also presents completely new challenges to the scientific concept and the theory formation of health care research we have not yet penetrated (…).” [4]
The distance ultimately means that in health and health services research, approaches that approach the health system in an exploratory manner and that - in contrast to experimental approaches - are not dedicated to testing hypotheses, but rather to generating new knowledge, are neglected. In this respect too, we agree with the authors of the memorandum just quoted: “Theory requires imagination, and this imagination often makes use of completely incalculable sources and suggestions, perhaps least of all the rigid framework of formalized hypothesis testing.” [4]
Association research – an untrodden path to new insights
The exploration of Internet transaction data could be helpful in many areas in the healthcare sector if they were available in the German healthcare context13. The question therefore arises as to which options provide transaction data that does not come from Internet applications but from business processes of statutory health insurance in Germany. The data known as “routine data” is also available in large quantities and has its strength in the area of morbidity and interventions, rather than in the area of the preferences and choices of the insured or patients. With them, lively research has emerged, which has enriched the picture of care by several aspects14.
What remains largely unaddressed, however, is the fact that caring for a large population essentially represents a huge natural experiment: millions of people suffer from diseases that are coded with thousands of diagnoses and are also exposed to thousands of different therapies in countless episodes. A large part of the treatments should not pose any puzzles because they depict exactly what the guidelines or the textbook require.
However, there should also be significant variances that arise in the real world because not all people react in the same way because they belong to genetically different phenotypes. These variances will show up in the data, although we cannot readily see the reasons for this due to a lack of data. Today we know that such effects can have a significant influence on the care process and ultimately also on the success of therapies. You should know a lot more about this.
In a large population, one will find that many patients fall into patterns of comorbidities that are expected based on medical knowledge. Metabolic syndrome is a common example of this. The decoding of the genetic control of cellular processes and the molecular basis of the control of cell functions is increasingly teaching us that individual genetic constellations and signaling processes are involved in sometimes very different diseases that have not previously been seen in connection. In large populations there is a chance of discovering such rare connections due to the great statistical power. This could also be used to develop a valuable instrument for researching biomedical connections. Transaction data from the GKV could help to support or reject biomedical hypotheses.
In addition, we can assume that applied therapies not only unfold their intended effects, i.e. affect the disease for which they are approved. On the one hand, their effect will be influenced by comorbidities, some of which are known from clinical tests, but some are not, because the number of people tested there is too small. On the other hand, however, the applied therapies are also likely to have effects that no one previously expected. This does not mean “off label use” in a few individual cases, but rather the broad effect in other disease states. For example, there could be an opportunity to use long-established drugs to treat conditions for which they were not originally intended. We should also know more about this, because the "repurposing" of drugs could possibly help progress in the search for therapies, where we are still groping in the dark today (e.g. dementia [8] [6]).
These outlined applications could be part of a broad-based association research, which – if you wanted to – could easily be described as a “big data” application in the sense mentioned above. Characteristics would be the origin of the data from transactions or routine processes and their very large number. The fact that we are explicitly not dealing with experimental research here, but with classic observational studies in which correlation and the control of confounding with statistical means are the methodological choice, is also formative. But why shouldn't it be possible to transfer the potential of Amazon analytics to health research?
Couldn't the algorithm for "Customers who bought this item also bought..." also serve well for the task "Patients who have disease X also have disease..."?
Such projects have already been launched elsewhere, including using a combination of clinical and genetic data. [7]
The rationale for such a program comes from the goal of developing new knowledge, which has always been the impetus for scientific change. Paradigm shifts have never occurred because an experiment could falsify a hypothesis, but because a new approach promised many new research projects and many new findings. [9]
Incidentally, all those who work in clinical research to banish error from the results can confidently adhere to the scientific understanding of critical rationalism. But there are many more paths to new knowledge that we should not block through methodological orthodoxy. We’re still a long way from having “Enough.” <<