In their letter, Verma and colleagues make the interesting point that possibly the age at first stimulant use × current age interaction effect found in our paper ^{Reference Groenman, Oosterlaan, Rommelse, Franke, Greven and Hoekstra1} might be influenced by our selection of patients. Including individuals with stimulant treatment duration longer than 12 months in our analyses, we found a protective effect of earlier age at first stimulant use on the development of substance use disorder (odds ratio (OR) = 0.95, Wald χ² = 13.78, P<0.001). Verma et al are concerned that we excluded patients with shorter treatment durations. However, when we include all individuals who ever used stimulants, we find the same effect (OR = 0.95, Wald χ² = 11.89, P = 0.001). Purely for illustrative purposes, we plotted the predicted probability of substance use disorder for the control group in Fig. 1. The figure shows that delay in the first age at stimulant use leads to marked increases in the probability of developing substance use disorder. In our article, we examined whether the effect of stimulant treatment depended on other factors (i.e. current use of stimulants, age at stimulant treatment initiation, age-adjusted duration of stimulant use and age-adjusted cumulative dosage), but found no other significant predictors than age at first stimulant use.

Fig. 1 Predicted probability of substance use disorders (SUDs).

Purely for illustrative purposes, we plotted the predicted probability of substance use disorder according to a generalised estimated equation model, that included age, gender and the interaction between age and age at first stimulant use. For healthy controls the model only includes age and gender. Below average age at first stimulant use: participants started before age 8.1 years; above average age at first stimulant use: participants started after age 8.1 years. Please note that predictions for healthy controls are not the product of the same model as prediction for stimulant groups.

Verma and colleagues refer to a meta-analysis, but provide the wrong citation. Recently, a meta-analysis on this topic was published. ^{Reference Humphreys, Eng and Lee2} Here no difference was found between treated and untreated patients in risk of developing substance use disorder (including alcohol, marijuana, cocaine and non-specific drugs) and nicotine use. Unfortunately, in this meta-analysis specific moderator variables such as age at first stimulant use were not taken into account, probably because of the relatively low numbers of studies to date that include such variables.

We thank the authors for discovering the mistake in the table, 9% should have read 59%.