Skip to Main Content


We have covered a lot of ground so far in this book, reviewing multiple methods ranging from medical research to automotive manufacturing to machine learning. Our goal is to improve, measure, and evaluate changes in healthcare delivery. But fundamentally, we are interested in whether a change in the care we provide benefits patients and is not merely associated with a benefit. Does the availability of board-certified intensivists improve the care of the critically ill at night?1 Does the introduction of an electronic health record (EHR) improve or harm the care of patients?2 How can we know when our interventions have actually made a difference to our patients?

Unlike other fields of research, healthcare delivery only occasionally lends itself to a randomized control trial. To randomize patients to one type of care-delivery strategy requires that we have equipoise, which is to say that we must be truly uncertain about the benefit of our intervention if we are administering it to only a fraction of our study population. It is hard to argue against efforts to improve the sterility of central line insertion3 or same-day primary care appointments.4 How would we respond if we were asked to enroll our aging and critically ill mother in a randomized study where, in one arm, she’d be cared for overnight in the intensive care unit (ICU) by a first-year resident and, in the other arm, she’d be cared for by a board-certified intensivist with at least seven years of post-medical-school training?1 Would we allow her to participate in that study? Healthcare delivery interventions that use randomization also require huge commitments of resources, including the staff, implementation, and leadership for two different care pathways, which could almost never be blinded to one another. But as healthcare delivery scientists, we have to remain open to the idea that what seems like the obviously correct path of care may turn out to be a very wrong path of care for patients. (See Chapter 1 for some examples.) Maybe, as it turns out, having a board-certified intensivist present in the hospital does not meaningfully change a patient’s care overnight and, as a downside, is incredibly expensive. Maybe we could have enrolled our mom in that trial, and it wouldn’t have harmed her at all.

Fortunately (and unfortunately), the care we provide in hospitals and in the outpatient setting is often random in nature. While unnecessary variation can be frustrating and wasteful, it does have an upside. Variation in care provides us opportunities to study what we do and attempt to draw true causal inferences from our care. For example, suppose that we are interested in whether the use of a transthoracic echocardiogram within 2 hours of admission with shock improves inpatient mortality. We could imagine that transthoracic echocardiograms are available only to patients admitted during the weekday business hours in our hospital. Essentially, knowing whether a patient comes into ...

Pop-up div Successfully Displayed

This div only appears when the trigger link is hovered over. Otherwise it is hidden from view.