Computational models have been instrumental in describing underlying cognitive and neural mechanisms of learning and decision making. The current thesis centers on the use of such models in order to better understand the processes leading to impaired decision making and learning in attention-deficit hyperactivity disorder (ADHD). Low signaling of the neurotransmitters dopamine (DA) and noradrenaline (NA) in ADHD are hypothesized to be the cause of the impairments, but the mechanisms leading to problems in decision making and learning is unclear.
All three papers of the thesis use the drift diffusion model (DDM) to explain mechanisms of simple two-choice decision making. Much of the popularity of the DDM was driven by the discovery that neural firing in monkeys performing decision making tasks closely resembled predictions made by models like the DDM, namely that clusters of neurons in monkeys accumulate evidence until reaching a decision threshold. Neuronal firing is sustained when decisions have to be maintained, suggesting decision neurons in monkeys both accumulate evidence and maintain choices. In paper 1, we tested if decision neurons in humans also maintain evidence, by comparing neural activity, measured with functional Magnetic Resonance Imaging (fMRI), during immediate and delayed response conditions. No brain region displayed activation across response conditions and difficulty levels consistent with predictions for a combined accumulation and maintenance region. Instead, our results supported results from previous studies on decision processes in humans, suggesting that evidence accumulation is performed across fronto-parietal regions. The results thus indicate a dissociation of evidence accumulation and decision maintenance in humans, contrary to reports on monkeys.
DA and NA are involved in several sub-processes of decision making. These sub-processes can in turn be linked to parameters in the DDM. Our aim in paper 2 was therefore to use the DDM to better describe the mechanisms leading to impaired decision making in ADHD, and how these mechanisms are altered by stimulant medication. We used a randomized placebo-controlled design to measure the effect of methylphenidate (MPH), a central stimulant increasing signaling of DA and NA, on decision making in adult patients with ADHD. Patients were also compared to age-matched healthy controls. We found that patients with ADHD responded prematurely and were slower at accumulating evidence than healthy controls, both on and off MPH. MPH increased the rate of evidence accumulation, but did not improve premature responding. Our results thus suggest a selective effect of MPH in ADHD improving evidence accumulation during decision making.
Instrumental learning involves an active choice process, but current reinforcement learning (RL) models usually describe choice processes in learning through a single parameter, which potentially captures several independent decision processes. In paper 3 we created a model to disentangle the decision processes; by describing an RL model (the RLDDM) that explains choosing in instrumental learning with the DDM. The RLDDM was validated by comparing fit to data across models with different expressions of parameters. The best-fitting model was further shown to recover generated data, suggesting that the model parameters captured the mechanisms they were intended to describe.
ADHD is associated with impaired learning. Despite strong connections between ADHD and DA on one side, and DA and learning on the other, few studies have directly used RL models to test the effects of central stimulants on instrumental learning in ADHD. Using data collected for an earlier study, we measured the effects of stimulant medication on instrumental learning in adult ADHD with the newly developed RLDDM. Results indicated that stimulant medication resulted in improved evidence accumulation, more deliberate choice processes, extended motor-processes, and slightly more conservative updating of beliefs from unexpected rewards.
Together, the current thesis implicate that central stimulants improve evidence accumulation in ADHD across both simple decision making and instrumental learning. The thesis further discusses the potential benefit of using computational models to understand cognitive mechanisms, and their alteration, in ADHD. Finally, the work presented herein highlights the importance of continuously improving computational models’ ability to describe observed behavior.