Abstract
When we build back better from COVID-19 we will need a better way of harnessing the data generated by health systems. Research methods and statistical analysis in medical journals are rooted in traditional methodologies such as observational studies and randomised controlled trials. But the world is changing, and already changed. It isn’t always possible to wait for a randomised controlled trial to judge the effect of an intervention, for example when we are considering the value of lockdown measures in a global pandemic. We might turn to modelling, as many countries did, but surely a better answer arises from using real data?
Health systems are overflowing with data, of interactions between patients and clinicians, of outcomes of care pathways, and of local and national practice and policy. The question is often how best to ‘clean’ those data and make sense of them outside the rigours of a traditional study design. As artificial intelligence, machine learning and wearables add to a data deluge, the answer is that each health system must optimise its analysis and use of data. To that end, Ben Goldacre and colleagues make a comprehensive proposal for how to bring NHS data analysis into the 21st century.1
These innovations create a challenge for policy makers. They also test medical journals since these analyses sit outside the usual methodologies that are within the competence and scope of journals. But journals must respond, because the data analyses will become more diverse, not less, and based on larger and larger datasets. Decisions and information will be required urgently, whether responding to the next global pandemic, climate change or a local dispute about service provision.
Another problem for medical journals is what to do about intervention studies that are not randomised trials. The answer usually is to decline to publish them. But what if a reliable dataset examines a question of acute interest, and is in a field where a randomised trial isn’t feasible or an observational study offers us the best answer so far? What do you do with those findings that might help a patient? The JRSM’s answer in the case of a tertiary centre’s extensive experience using cognitive behavioural therapy for chronic fatigue was to publish the paper.2 The answer from the study isn’t definitive but it is suggestive.
And that’s the crux of the data argument: modern data analysis may not be the final word but it may be good enough to be helpful, and preferable to an answer based on guesswork. Where a randomised trial is feasible, of course, it should be done. But how do you quickly answer a question about not treating fever in early COVID-19 or social prescribing for population health in a lockdown?3,4 Analysis of health service data can help.
If we accept the centrality of data to our present and future, in our rush to embrace innovative data analysis, we must not forget the importance of softer skills such as empathy and learning from history.5,6 In the final reckoning, data are only as good as the people, or machines, who use them.