In this study, we aimed to identify clinically relevant predictive factors for pharmacological ADHD treatment effects in children and adolescents in a clinical setting. When performing unadjusted multinomial logistic regression without validation, baseline higher SNAP scores for inattention, combined, and the total score, and being treated in the Gotland region compared to the Västerbotten region as well as being treated in the Västerbotten region compared to the Stockholm region, increased the odds of being a responder compared to non-responders. Baseline higher SNAP scores for hyperactivity/impulsivity, combined, and the total score, as well as stating ADHD medication at three months, and being born in the second tertile compared to the third tertile, increased the odds of being an intermediate responder compared to non-responders. However, none of the significant variables could explain more than 1.8% of the variation in the model. Consequently, even though crude estimates were significant, we could not identify any variables predicting pharmacological treatment effect.

Our findings align with prior research indicating that more severe ADHD symptoms tend to correlate with improved treatment outcomes [16]. In contrast, some studies suggested that more severe ADHD symptoms might predict poorer treatment outcomes [11, 14]. The discrepancies in findings could be attributed to variations in measurement criteria and defining outcomes. We applied a strict definition of responders (≥ 40% SNAP score reduction) [36]. One-third (33%) in our cohort were classified as responders, which in comparison to other studies, is a low number. [11, 14]. However, reducing the responder definition to ≥ 30% did not yield significant changes in the results. Neither did analyzing SNAP-IV scores as a continuous variable. Also, there is no accepted definition of clinically significant treatment response [10].

In our data set, the Gotland Region demonstrated a significantly higher number of responders compared to the other participating regions. Nevertheless, only 19 participants from Gotland were included in our study. Consequently, drawing any definitive conclusions from this result is not feasible.

To the best of our knowledge, our study is the first to examine blood pressure, heart rate, relative age, pharmacological treatment initiation month, and psychotic-like experiences in the context of response rate of ADHD medication. Blood pressure and heart rate are closely intertwined with the sympathetic and parasympathetic nervous systems which are affected by many psychiatric disorders [37]. Psychotic-like experiences has been shown to strongly correlate with clinical psychotic disorder [33], which might indicate severe psychiatric illness. The relative age effect refers to the phenomenon where individuals born closer to the cutoff date for school start demonstrating advantages over their younger peers. Studies have shown that the youngest children in a school class are more likely to be diagnosed with ADHD [38, 39] or receive ADHD medication [40]. Thus, we hypothesized relative age might affect our outcome. Surprisingly, in our study, tertiles of birth month were evenly distributed throughout the year and had only a minor effect on treatment outcome.

To further investigate the predictive potential of our variables, we applied Machine Learning algorithms to our dataset [41]. The ROC curve for the Boostrap Forest model was significant within the training dataset but could not be replicated in the independent validation dataset. This means that none of our variables, including those identified as significant through Multinomial Logistic Regression analyses, could predict treatment outcomes when subjected to novel data.

Our results accord with earlier studies showing that despite extensive research, the evidence regarding factors associated with outcome of pharmacological treatment remains inconclusive.

This study’s primary strength is that it expands the scope of current research on factors linked to pharmacological treatment outcome through its contribution with an extensive sample size of ADHD patients derived from a representative clinical cohort, which mirrors real-world evidence. We recruited children and adolescents from three Swedish regions and introduced clinically relevant predictive factors, some never studied before.

Moreover, we performed both conventional Multinomial Logistic Regression and Machine learning techniques to ensure the robustness of our findings. Validation with Multinomial Logistic Regression and Boostrap Forest resulted in identical results, further bolstering our conclusion.

Our study underscores the challenges with incomplete data, frequently encountered in clinical research. This was particularly evident regarding information on current ADHD medication. A considerable number of individuals lacked information on ADHD medication at three months, categorizing them as missing cases. However, according to local clinical guidelines, children discontinuing pharmacological treatment were excluded from the psychiatry units. Thus, presumably, all children in the cohort, except for those stating No medication, had medication at three months. This likely resulted in an underestimation of individuals with ADHD medication, potentially influencing our models.

Varying attrition rates and loss to follow-up might introduce a bias. However, there were no significant differences in baseline characteristics between children with and without a three-month follow-up. Importantly, when we performed sensitivity analyses with more complete data, all our results were sustained, indicating the robustness of our dataset.

Unfortunately, we lacked information on medication compliance, a factor that could influence response to treatment. Neither could we evaluate the influence of medication dosage due to insufficiently registered data. Also, the Ethical permit did not allow access to medical records. Studies have demonstrated that increasing the dosage of stimulant medication enhances pharmacological efficacy [42]. However, Vallejo-Valdivielso et al. did not find a difference in MPH dose between responders and non-responders [11].

Information on eligible patients declining participation was lacking, thus participation rate could not be calculated. Lastly, the questionnaires in the study are all parent-rated, which could have distorted presenting symptoms in our cohort. Including clinician- and teacher-ratings would have been beneficial.

Our negative findings are important as an increasing number of children and adolescents are treated with ADHD medication, even though it is not well elucidated if a child will benefit from treatment.

Our results support the compiled evidence that to date, distinct predictors for the treatment effect of ADHD medication have not yet been revealed [22].

There are several possible explanations for the challenge of finding predictors of treatment outcomes.

The inconsistency across studies giving rise to precarious comparisons, implies that the factors influencing pharmacological treatment effects may be as multifaceted as the ADHD diagnosis itself. Further, we might not measure the right things [43]. Despite the use of validated questionnaires, our findings indicate that the tools used may not fully capture the fundamental challenges faced by children and adolescents with ADHD.

Another potential explanation may be that the medication primarily addresses factors beyond the DSM-5 diagnostic symptoms of ADHD, which are currently not measurable.

Although more studies on the influence of external factors, such as e.g. school environment, and traumatic experiences [44] are needed, it may be more successful in addressing underlying transdiagnostic neurophysiological [45]and biological features [46]as possible predictors of ADHD medication response.