Implications of prediction
Having conducted policy research for the Department of Health on
predictive modelling, I remain concerned about the unexamined implications
of their use. Yes, the algorithm does help sort through health databases
to identify high-risk individuals. But what actions can properly be taken
with that information by the NHS?
Have patients consented to be profiled; indeed what would they make
of using ethnicity as a criterion? Should we expect people identified to
be at high risk and high utilisation to be called in by their GP for a
'chat' about their potential future use of the health care system?
There is a relatively small administrative step between determining
what resources are appropriate for what level of risk as determined by
these sorts of algorithms and deciding what to do when a patient runs over
their apparent allocation, especially since it is a founding value of the
NHS precisely not to do this. Perhaps, all patients will want to be
profiled in order to ascertain their fair share of resources for their own
healthcare. And then what happens to the relationship between the NHS and
The authors speak of designing appropriate interventions. Will
patients feel pressured or coerced into participating? And of course once
designed, these interventions had better work as failure will point once
again to the features of the patient and not the system.
As I reported in my work (which for its own reasons the BMJ chose not
to publish...) there are considerable benefits and risks from using
predictive algorithms in the NHS. The social context of their use and
misuse need examination particularly in respect of data protection and
confidentiality. My concerns here are exemplified by the capabilities
reported in this paper.
The author conducted policy research for the Department of Health on predictive modelling.
Competing interests: No competing interests