I’ve been thinking lately about how the data we have, which informs policymaking around pop health, can & does exclude people who do not have insurance, who forego health care, who are underbanked...

& training datasets for predictive AI reinforce those absences/exclusions.
For example, much of the data on prevalence of XYZ conditions is based on records of patients who used health services. But, given the layers & layers of rationing, there are people who cleared hurdles. They can be qualitatively different than those who did not.
If we wish to do research that informs policymaking & priority-setting for a universal health care system- financing, non-financial facets of access- we need to expand the universe beyond those who currently have access.
For example, ~30 million people in the US remain uninsured, & many more are underinsured. This constrains their use of preventive care, & lessens the likelihood of EHR detailing their health status trajectory over time. We cannot forecast the population needs based on this data.
(Or maybe we can. But the assumptions should be made explicit, & the ‘missing data’ should be acknowledged)
So, the closest thing we have to an assertion of healthcare as a right is regulations that prevent hospitals turning away patients in the ER. Among this pop are the un(der)insured, un(der)employed, & those who currently do not qualify for subsidies for ACA marketplace plans...
Absent preventive care, a ‘medical home’, etc, these patients’ records may be fragmented & only reflect episodic care for acute or untreated/undx’d chronic conditions.

What do these data teach an EHR-based AI when they are part of a training dataset?
And what does their absence teach an AI?
Recently, I have been reading about autopsy as population health surveillance, & the financial incentives for local ME offices to ‘capture’ as many bodies as possible- disproportionately impoverished, Black, & indigenous.

It’s giving me a lot to chew on.
I see parallels with the below- in death as in life https://mobile.twitter.com/Arrianna_Planey/status/1094276651012292609
Anyway, just some thoughts over coffee
You can follow @Arrianna_Planey.
Tip: mention @twtextapp on a Twitter thread with the keyword “unroll” to get a link to it.

Latest Threads Unrolled: