Biostatistics

The statistical meaning of power is the probability that a future study will turn out successful, given some anticipated level of (drug) effect. Regulatory authorities usually require at least 80% power of a planned study, and up to 90% is common. After a study has been performed, the power is equal to 0 or 100% – either the study was successful or not.

To be frank, a statistican cannot be called in too early. Statistical thinking from early on in study planning, or even when planning what studies to run, is a great tool for optimizing the benefit, time lines and budget of any project

Are statisticians just number crunchers? 

Some people think of statisticians as number crunchers, who will help them sort out the big mess of data they ended up with. We certainly like doing that, but with statistics support already in the study design phase the mess is avoided altogether, and the data analysis gets so much smoother and quicker. We have all suffered the sight of data beyond rescue, where the study happened not to measure that optimal variable,  missed to include the most informable control group, or failed to capture the scientific question because the RWE data had a different structure than anticipated.

The benefit of calling in statisticians early

While of course performing a lot of data analyses, our statisticians spend just as much of their time in strategic planning phases with study teams. They help evaluate parallel-group or cross-over designs, different versions of endpoints, power and sample size for different study designs, practical aspects of data collection and much more, all towards the most efficient and informative connection between a research question and the resulting output of a study.

So, don’t hesitate to approach your statistician early on. The shortest chat can sometimes make all the difference to your final results.

Certainly not! Sometimes discussions start with a tentative caveat: “Perhaps this is ridiculously easy and a waste of your time”. Our experience is, even the seemingly minor question may have delicate aspects. The waste of the statistician’s time, and yours, is when we miss that small detail which may have a huge influence at the end.

So, isn’t statistics all just lies? Maybe it sometimes can be in the wrong hands. It is true that us humans tend to attend to facts that support our preconceived opinion, and if we just keep collecting the odd results that support us, we will eventually end up with a whole pile of evidence for justabout anything, without telling the full story that 99% of similar investigations showed the opposite.

The Statistical Analysis Plan

The Statistical Analysis Plan, or SAP, is a detailed plan of all statistical analyses which will be carried out in a study. The sophistication of the SAP is that it is written before anybody knows what the data results look like. This is an important feature, as otherwise, the data analyst may subjectively choose between different statistical methods until he or she finds one that gives the desired results. Even just the fact that the option to do so exists, diminishes the data integrity of the study, meaning the statistical results are less valuable as evidence at the end of the day. The core value of the study is at stake.

In clinical studies, which are highly regulated by authorities, the SAP is of vital importance. Every detail of statistical analysis is neatly planned before data breakage, even though methods can potentially be conditional on data formats. In this way, the interpretation of results is clear-cut. All parts understand the overall study results. There is no possibility to show only favorable results to the regulatory authorities, and hide selected bits and pieces.

In general, statisticians work according to statistics ethical principles whether or not they are in the world of drug development. We see statistical analysis plans, sensitivity analyses, interpretational discussions and thorough review of research in all fields of data analysis.

This question will get very different answers depending on who you ask, but here is an attempt to a fair summary, from the perspective of a statistician.

Statistics

The old subject area of statistics has a few related pillars.  Statistics aims to translate a question into quantitative items which can be measured  in a practical and doable way, while optimizing the amount of information gathered and minimizing the uncertainty of the results. Are 10 year old girls taller than 10 year old boys (hypothesis testing)? How tall will my daughter be at the age of 10 (prediction)? What are the heights of boys and girls in this school class (data summary)? After assessments have been done (or based on existing data), statistics is the art of interpreting data to answer the original question, and perhaps also new ones. What can we conclude from the data? In particular, does this data suggest conclusions beyond the actual dataset? Measuring heights of a sample of 10-year olds, what can we say about 10-year olds in general?

Artificial Intelligence (AI)

Artificial Intelligence (AI), is a recent word for the overlapping field of using data for prediction. AI is casually thought to be based on very high dimensional data, like the properties of all mobile phone calls in a country, or the contents of the world’s collected google searches. With huge amounts of data being collected in the society, a lot of insights are right in front of us. If we can decode it. The vast data flows offer new types of challenges technically, but also on the level of interpretation – the old pillar of statistics.  Are mobile phone positions representative for people locations at all (e.g. Covid -19 tracing)?

Machine Learning

Machine learning is a set of prediction methods commonly used in AI, but not specific to AI. For a statistician, machine learning is one out of several tools in the box.

So, there is plenty of overlap between AI and statistics. Some statisticians work with AI, and AI experts always use more or less statistics concepts. The statistician may or may not put more effort into drawing sharp conclusions. Can we say that 10 year old girls are taller than 10 year old boys? The AI expert will often focus on deriving a most likely prediction. What google search is the writer about to type?