OBJECTIVES : Since the turn of the 21^st century, average health has improved around the world. At the same time, it has been suggested that many high-income countries are practicing “flat-of-the-curve” medicine, where average incremental gains from additional spending have become very small. While studying the trend in incremental QALYs reported in cost-effectiveness analyses (CEAs) would directly answer this question of declining marginal gains, we study here a potential behavioral response by researchers seeking to establish the value of interventions. Specifically, we explore whether utility weights used in CEAs show a trend toward more significant digits.

METHODS : To assemble a dataset of health-state utility values, we collected all utility weights from CEAs reported in the Tufts CEA Registry between 2002 and 2017, along with corresponding study characteristics. After excluding health states representing death or perfect health, our dataset contained 15,549 utilities. We defined “high precision” as a utility weight with three or more significant digits. We fit a logistic regression model of the probability of a utility being reported with high precision each year while adjusting for disease area and funding source.

RESULTS : Significant digits in utility weights are increasing over time. Overall, 15.7% of utilities in 2002 had high precision compared to 45.2% in 2016. Each year since 2002, the probability that a health-state utility weight was reported with three or more significant digits increased by 8.3% (95% CI: 7.2-9.4%, p < 0.001). This conclusion was robust to sensitivity analyses in which we redefined “high precision” as having four or more significant digits, used study-level data, and refit the analysis separately by sponsorship.

CONCLUSIONS : There is a trend toward using health-state utilities with more significant digits. New methods in utility elicitation allow for the appearance of greater precision, but their use in CEAs must maintain a connection to meaningful interpretation in a clinical context.