Politics, ideology, and bias in bad data

I was recently granted access to a large database. A comment I made from a microphone at a healthcare conference created a dialogue extending beyond the 45 minute presentation and here I find myself--staring in disbelief at a modern dataset with limited applicability to the complexity of health outcomes at the patient or community level.

Typically in data scarce environments I either provide insights to help curate data for information or construct survey instruments based on robust research methodology and frameworks. On rare occasions I am presented with data I had zero hand in generating. I thought it might be interesting to share a few of the top of line concerns this creates--dramatically edited down from what is typically an hour long conversation.

Race as a political, social or biologic construct

I am looking at the demographic variables you have selected and I am typically (9 times out of 10) stumped at questions of ethnicity or race. Many of my data savvy colleagues will join me in befuddlement. There is only one race. There are not any genetic markers that define race. No this isn't a kumbaya moment, it is a fact. Race--The Power of An Illusion confronts the complexity head on. I wrote a bit about it here, When A Symptom Becomes the Disease with additional resources.

What information are you actually trying to measure? Muscle mass? Enzyme levels or variations in biochemistry? Perhaps you are using race as a substitute for social determinants of health such as educational levels, transportation, housing, access to clean water or nutritive food sources--to name a few. Be specific. I am unable to detect meaningful differences when variables are weak, ill-defined, or lack meaning.

What is the definition of mental health/illness, depression, or psychiatric illness?

We appreciate the emerging role of the patient in the healthcare conversation. Patient reported outcomes or PROs are a hot commodity in the buzzword zeitgeist. But what do we mean when we validate PROs to be used in discussions of value from a patient perspective? The Politics of Mental Health is a recent public lecture from the London School of Economics and Public Policy serving as a living example of the unsettled nature of how we define, diagnose, and treat patients with mental illness.

At the intersection of the personal and the political, we explore the relationship between mental health and economics, politics, and society at large. Is it even possible to distinguish between mental illness that derives from an individual’s physiology or childhood experience and that which has broader social or political causes? Why do particular mental illnesses appear to characterize certain eras? Could social change limit the spread of mental illness in contemporary society? 

Depression and feelings to adequately describe individual perceptions are often allocated into intentional states. What if accuracy requires not only clinical measures but a phenomenology of depression? Not only clinical symptoms but how an individual interacts with the world--interpersonal and community engagement? I challenge a single checked box or series of standardized questions to capture the nuance necessary to evaluate endpoints or clinical trial outcomes designed to capture efficacy.

How does mindfulness impact bias in survey responses?

Should we measure mindfulness or the values we attribute to advance arguments for respective positions in advancing educational goals or clinical research? For example, continuing medical education professionals measure healthcare provider knowledge, competence, or "behavior" before an educational intervention and following completion of the activity. Few are willing to apostatize the effectiveness of educational frameworks or specific interventions funded by industry. Better data can be collected and curated for insights objectively aligned with actual gaps at the point of care if perhaps we measure mindfulness to reveal influences such as social biases, vocational wellbeing, and emotional empathy.

Interested in the topic? Follow more conversations at Data & Donuts.