Research project Treatment methods for young people with self-harming behaviour: statistical method development for evaluation of long-term effects, suicide risk, and health economic aspects
Based on a novel register database of young people who self-harm, the project develops customised statistical methods to estimate causal effects, compare different treatment approaches, and evaluate the long-term cost-effectiveness of treatments.
The project analyses the long-term effects of self-harm behaviour in adolescents, with regard to treatment outcomes and costs, suicide risk, and living conditions. As part of the project, a comprehensive register database is being built, containing data on young people who have sought care for self-harm behaviour from ten national quality registers, as well as data from a randomised controlled trial (RCT). The database also includes information on the adolescents' parents and siblings. The project develops statistical methods specifically adapted for analysing and drawing causal conclusions from combined data collected through both an RCT and an observational study.
Randomised controlled trials (RCTs) are the gold standard when studying the effects of different treatments, due to their high validity, as the principles governing allocation to treatment groups are known and controlled. At the same time, they are often costly and time-consuming, and strict inclusion/exclusion criteria can make it problematic to generalise the results. A major advantage of register-based studies is that they facilitate follow-up over time, minimise the burden on patients, and can be conducted at lower cost for larger, more representative samples. However, there is a risk of bias due to so-called confounders that affect both allocation and outcome — for example, when the severity of a condition both makes it more likely that a patient receives a particular treatment and simultaneously influences the effect of that treatment. Statistical methods that ensure valid effect estimates by accounting and adjusting for confounders and other sources of error have been developed within the field of causal inference.
The project examines the long-term effects of two self-harm regulation treatments (IERITA and DBT-A), in comparison with treatment as usual, and investigates which patient groups benefit most from each treatment approach. DBT-A is an established but resource- and time-intensive treatment, while IERITA is a shorter internet-based intervention with the potential to reach a broader population. The effects and costs of the different treatment options will be evaluated through health economic analyses.
The project involves a number of statistical methodological components, including evaluating different methods for designing the matching of controls, ensuring that the necessary assumptions for causal inference are met, that bias is minimised, and that the efficiency of the study is maximised. The project will develop methods for effect estimation adapted to combined data, evaluate Bayesian estimation methods, and compare machine learning models for the prediction of suicide attempts and suicide. A further aim is to define cost and outcome measures for the health economic evaluations, as well as tailored analytical methods.