Defining and evaluating the Hawthorne effect in primary care, a systematic review and meta-analysis

Abstract

In 2015, we conducted a randomized controlled trial (RCT) in primary care to evaluate if posters and pamphlets dispensed in general practice waiting rooms enhanced vaccination uptake for seasonal influenza. Unexpectedly, vaccination uptake rose in both arms of the RCT whereas public health data indicated a decrease. We wondered if the design of the trial had led to a Hawthorne effect (HE). Searching the literature, we noticed that the definition of the HE was unclear if stated. Our objectives were to refine a definition of the HE for primary care, to evaluate its size, and to draw consequences for primary care research. We designed a Preferred Reporting Items for Systematic reviews and Meta-Analyses review and meta-analysis between January 2012 and March 2022. We included original reports defining the HE and reports measuring it without setting limitations. Definitions of the HE were collected and summarized. Main published outcomes were extracted and measures were analyzed to evaluate odds ratios (ORs) in primary care. The search led to 180 records, reduced on review to 74 for definition and 15 for quantification. Our definition of HE is “an aware or unconscious complex behavior change in a study environment, related to the complex interaction of four biases affecting the study subjects and investigators: selection bias, commitment and congruence bias, conformity and social desirability bias and observation and measurement bias.” Its size varies in time and depends on the education and professional position of the investigators and subjects, the study environment, and the outcome. There are overlap areas between the HE, placebo effect, and regression to the mean. In binary outcomes, the overall OR of the HE computed in primary care was 1.41 (95% CI: [1.13; 1.75]; I2 = 97%), but the significance of the HE disappears in well-designed studies. We conclude that the HE results from a complex system of interacting phenomena and appears to some degree in all experimental research, but its size can considerably be reduced by refining study designs.