Decoding Risk: Making Sense of Predictive Data Analytics

Abstract

As the data phenomenon has grown, researchers have increasingly recognized the need to investigate algorithmic and data-driven technology, practices, and culture. While a number of studies have investigated the process of designing algorithms or examined the contours and consequences of algorithms themselves, more research is needed that details the ways in which users of data-driven technology make sense of algorithmic output and the conditions under which these processes vary. This paper explores how users derive knowledge from predictive algorithms in medical contexts. Drawing upon interviews and observations of a neonatal intensive care unit, I examine how clinicians come to know things about patients and disease. In particular, I focus on their use of data-driven predictive monitoring technology designed to forecast the onset of infection. Although developers intend the technology to function as an early warning system, clinicians do not formulate knowledge from the technology as intended. Instead, I find that rather than treating this data as deterministic indications of disease, clinicians engage in a negotiation between experience and intuition, on the one hand, and a doctrine of demonstrable evidence on the other. This results in a series interpretive processes that I call “conditioned reading,” “accumulative reading,” and “retroactive reconditioning.” I suggest that these particular processes take place in response to an institution entrenched in a culture of evidence based medicine. I conclude by arguing that fully theorizing the social implications of data analytics will require researchers to investigate the role of institutional contexts.