Study exposes hidden bias in digital alcohol intervention trials

<p><strong>Background:</strong> Alcohol consumption is a significant global public health issue, causing approximately 2.6 million deaths in 2019, with 1.6 million attributed to noncommunicable diseases (NCDs) such as cardiovascular diseases and cancers. The World Health Organization emphasizes reducing alcohol intake as a key strategy for addressing NCDs and contributing to the attainment of the United Nations Sustainable Development Goals. In Sweden, 75% of the adult population consume alcohol and 40% are classified as risky drinkers according to national guidelines. The societal costs of alcohol-related consequences in Sweden amount to approximately 9 billion euros annually. There is, therefore, a pressing need for health promotion and disease prevention interventions targeting alcohol consumption. Digital interventions have emerged as promising tools to support individuals in reducing alcohol consumption. These interventions can reach large populations at low costs, but their long-term health benefits and cost-effectiveness remain uncertain. Health economic evaluations, including cost-effectiveness analyses, are crucial for assessing the long-term health benefits of interventions. These evaluations balance costs against health outcomes and integrate data from various sources, including randomised controlled trials (RCTs). However, the validity of RCT results can be compromised by methodological factors, such as research participation effects (RPEs). RPE is an umbrella term that refers to changes in behaviour, attitudes, or beliefs of individuals that occur simply because individuals are participating in a study. These changes result from the unique context of research, where participants may react unexpectedly to tasks they are invited to do, influencing their behaviour instead of the treatment provided. One example of how RPEs can be induced is by using waiting list control group designs. There is a small growing body of literature providing evidence that using waiting lists may lead to consequences for participants. However, this remains an underresearch area.</p><p><strong>Aim:</strong> The overall aim of this thesis is to investigate how RPEs may bias estimates of effects in trials of digital alcohol interventions, and how such biases propagate when simulating long-term outcomes.</p><p><strong>Studies:</strong> Five studies were conducted to address the aim of this thesis. In Study I, an individual level simulation model was developed and applied to investigate long-term outcomes of a digital alcohol intervention, including the incidence of 10 alcohol-related diseases, quality-adjusted life years (QALYs), and costs. The results showed that access to the intervention could lead to fewer disease cases, in particular alcohol-related liver disease and liver cancer, more QALYs, and lower costs compared to referral to websites with information on alcohol and health. A systematic review was conducted in Study II to investigate which RPEs have been studied in alcohol-related research. The systematic review identified studies that investigated RPEs stemming from four design choices: informed consent, assessment, group allocation, and waiting list. In Studies III and IV, waiting list designs were investigated further, and the effects of being allocated to a waiting list on alcohol consumption were estimated in an RCT. The results from Study III showed that participants were not neutral to being allocated to a waiting list control group in an alcohol intervention study, rather, being told that access to the intervention would be delayed had a negative impact on participants. Results from Study IV provided evidence that participants allocated to a waiting list control group were one month later more likely to report higher alcohol consumption than participants allocated to an intervention group, however, considerable uncertainty remained regarding these estimates. In Study V, the impact of four RPEs on simulated long-term outcomes of a digital alcohol intervention were quantified by accounting for the impact of RPEs on the estimated intervention effect. Results showed that ignoring RPEs can both under and overestimate long-term outcomes, potentially affecting policy decisions.</p><p><strong>Conclusions:</strong> Disseminating digital alcohol interventions can help reduce alcohol-related diseases, such as liver disease and liver cancer, improving public health cost-effectively. While trial design choices may introduce biases, these interventions remain cost-effective if biases are minimal. However, modest intervention effects make these biases more concerning and should be carefully considered. Design choices lead to RPEs which may not only bias estimates of effects, but may also have negative impacts on participants’ behaviour. The use of waiting lists is one example, and given their potential harm, it is crucial to explore ways to mitigate their negative impact and consider alternative control group designs.</p>