Uncategorized

The Data Response Curve (In Honor of the Dose Response Curve)


Data Response Curve

All medical students learn about the dose response curve in pharmacology lectures. The dose-response curve informs us of how we should dose a medication in the context of its efficacy and its toxicity. Too little medicine won’t have the desired effect, and too much medicine can be toxic.

In the era of digital health, data have become the new “big pharma,” and we are facing the emergence of a data-response curve in which access to too little data is inactionable, and access to too much data can be overwhelming. Digital health devices abound today, and has enabled quantification of nearly every health and wellness metric imaginable. Sadly, in our exuberance about these new sources of data, we often conflate “more data” with “better data.”

In the era in which data have become the booming commodity of exchange in healthcare, we describe an emerging data-response curve. Large data sets can be at best clarifying or at worst self-contradictory. Too little data on the data-response curve, as with medication dosing, can be insufficient for effective action or decision-making. Too much data can be toxic to the user such as the physician, leading to poor decisions or worse, to analysis paralysis.

We live in a world of exploding data, and we need to be thoughtful. Medicine is a people business in need of data, not a data business on need of people. In reductive form, all humans make decisions based on inputs (data) from their environment and on individual analysis of the data in the form of non-linear, non-quantifiable perception. When we as doctors, policy-makers, or simply as human beings receive these inputs, one of three things happens: 1. We make a decision to do something (for example a treatment decision based on abnormal data), 2. We make a decision to do nothing (for example, normal data which we believe requires no action), or 3. We need more data in order to make an informed decision to do item #1 or item #2. More does not necessarily equal better data. More data is simply more.

Better data are actionable data wrapped in the context of the patient and the patient’s condition. Imagine each piece of objective data connected to concurrent subjective data, and surfaced in the context of a specific condition relevant to the patient. HealthLoop enables patients to generate contextual objective data married to their subjective symptoms and served to a clinician in an actionable context. High signal and low noise are the digital health equivalents of on the dose-response curve of a favorable therapeutic window.

See where some folks live on the Chart. Where do you live?

10 replies »

  1. Jordan: Thanks for this excellent summary of some of the quandaries of health care information — and the only plausible glimmer of a solution.

    May I suggest the following list of desiderata?

    • Contextualizing data assures relevance.
    • Sufficiency of relevant data assures thoroughness.
    • Clarity and simplicity in the presentation of relevant and thorough data enables good decision-making
    • Tools that easily translate good decisions into corresponding actions are the keys that unlock the value of all the data we have accumulated.

  2. Great analogy — this helps health care professionals and their patients understand that getting the right information at the right time will mean better health.

  3. Fantastic post. So many third parties have oceans of data. Not only do they have no idea what to do with it, they CAN’T do anything with it. CMS comes to mind.

  4. Jacob,

    I’m glad you exist! This is a great refinement. I have come up with a new term which you will hear more about later “theta data” which is information that is actionable. Stay tuned.

  5. Thank you. My goal is to begin to refine what information is constant, what information is variable and what information is actionable.

  6. Jordan: I love the graphic and the message. But I disagree with how some (many?) might interpret what you’re saying without a small modification. My adjunct would be the addition of “presentation” to your headline message.

    “The data PRESENTATION response curve.” Is a better expression of what (I think) you are aiming for.

    What we find in the nooks and crannies of “big data” may in fact be some of the most valuable insights that will help us make more accurate diagnoses, or make better treatment or diagnostic decisions.

    More data IS just “more data” unless we can do something with it. But I think that we can. Our tools need to consume the data, help us perform analyses – and guide us toward the best options. This means that the PRESENTATION of “big data” needs to be clean, clear and easily understandable. It DOESN’T mean that the data itself is too big. I would argue that more really IS better on the back-end, while on the front-end, LESS is better.. John Maeda’s “laws of Simplicity” come to mind – along with Daniel Cabrera’s adaptation of the simplicity model for emergency medicine.

  7. Good post. You’re right.

    I started my white collar career in a forensic radiation lab in Oak Ridge in the 80’s, where most of the data coming out of the apps I wrote were to serve as evidence in dose/exposure/contamination litigation. I’ve been audited right down to my rounding algorithms. I have ever since been appalled at the cavalier acceptance of crap data. The new “Big Data” fad in healthcare, with its ever-present problem of “error propagation” resulting from mixing data of varying accuracy pedigrees, may well only amplify the problem.

  8. Very insightful post. I have found this to be true with all forms of information, not only data. Information overload is information overload, but try explaining that to somebody selliing information ..