In their scientific work, researchers rely on the data available to them to draw conclusions based on empirical results. However, when values are missing the interpretation of results can be difficult especially if it remains unknown what caused the dropout to occur. With regard to studies in clinical psychology, this could prove to be a critical point:
How effective is the treatment really? What happens to those participants who left the study leaving no data to indicate their level of mental health?
Many times when dropout occurs in a study analytical methods are applied that assume that the missing is either random or that all relevant covariates have been integrated in the model (missing at random assumption, MAR). However how can we know if this is correct? As an alternative and an addition to these classical models, NMAR (not missing at random) models have been developed. These models provide us with the opportunity to investigate dropout and understand it’s mechanisms in more depth. This could provide useful especially with regard to studies that focus on the effectiveness of specific interventions and at the same time report rather high dropout rates. One example for such cases are studies focusing on psychological internet interventions, where participants stop to fill out in-treatment questionnaires for unknown reasons. In our example, we investigated a CBT- based internet intervention that targeted depressive symptoms in 483 participants over a course of twelve weeks. In total 222 participants stopped completing the in-treatment questionnaires prematurely.
If we ask ourselves why these participants dropped out, we can come up with different answers: Possibly participants dropped out because their last observed score of depression was rather high. Other reasons may include initially negative attitudes towards internet interventions or a pattern of deterioration during treatment.
Depending on the answer, we may come to different conclusions regarding treatment outcome for dropped out participants. In our study, we tried to shed some more light on dropout mechanisms. Our results support the following findings: While dropout is related to deterioration, it is also related to fast improvement when initial impairment is low. Participants with low impairment and fast improvement seem to be able to maintain their positive treatment outcome in long term. In addition, a group of participants was identified that experienced no change during treatment, but did not drop out. In the same line as deteriorating participants, these participants may profit from treatment selection or adaption. Compared to a classical model the mean change estimated by applying a NMAR model was lower. This indicates a potential overestimation of the treatment effect when the classical model is applied and it underlines the necessity to further investigate and consider mechanisms of dropout. In our study, we were also interested in predictors of dropout and found attitudes towards internet intervention, age and level of mental and physical health to be relevant patient variables. Further research in this area could help to establish internet interventions that are sensitive to baseline patient characteristics as well as symptom development. While the application of feedback tools have gained attention in face-to- face therapy such approaches are still not as developed in internet interventions, although this is a setting where they may be especially needed.
Read the full paper: Arndt, A., Lutz, W., Rubel, J., Berger, T., Meyer, B., Schroder, J., Spath, C., Hautzinger, M., Fuhr, K., Rose, M., Hohagen, F., Klein, J. P., & Moritz, S. (in press). Identifying change-dropout patterns during an Internet-based intervention for depression by applying the Muthen-Roy model. Cognitive Behaviour Therapy. doi: 10.1080/16506073.2018.1556331
Photo by: AUM OER
Pictured: Alice Arndt & Julian Rubel