Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2020 Aug 1.
Published in final edited form as: Psychopharmacology (Berl). 2019 Jul 13;236(8):2501–2512. doi: 10.1007/s00213-019-05316-x

Modeling The Effects Of Methylphenidate On Interference And Evidence Accumulation Processes Using The Conflict Linear Ballistic Accumulator

Alexander Weigard 1,2, Andrew Heathcote 3, Chandra Sripada 1
PMCID: PMC6697566  NIHMSID: NIHMS1534632  PMID: 31302719

Abstract

Rationale:

Although methylphenidate and other stimulants have been demonstrated to improve task performance across a variety of domains, a computationally rigorous account of how these drugs alter cognitive processing remains elusive. Recent applications of mathematical models of cognitive processing and electrophysiological methods to this question have suggested that stimulants improve the integrity of evidence accumulation processes for relevant choices, potentially through catecholaminergic modulation of neural signal-to-noise ratios. However, this nascent line of work has thus far been limited to simple perceptual tasks and has largely omitted more complex “conflict” paradigms that contain experimental manipulations of specific top-down interference resolution processes.

Objectives and Methods:

To address this gap, this study applied the Conflict Linear Ballistic Accumulator (LBA), a newly proposed model designed for conflict tasks, to data from healthy adults who performed the Multi-Source Interference Task (MSIT) after acute methylphenidate or placebo challenge.

Results:

Model-based analyses revealed that methylphenidate improved performance by reducing individuals’ response thresholds and by enhancing evidence accumulation processes across all task conditions, either by improving the quality of evidence or by reducing variability in accumulation processes. In contrast, the drug did not reduce bottom-up interference or selectively facilitate top-down interference resolution processes probed by the experimental conflict manipulation.

Conclusions:

Enhancement of evidence accumulation is a biologically plausible and task-general mechanism of stimulant effects on cognition. Moreover, the assumption that methylphenidate’s effects on behavior are only visible with complex “executive” tasks may be misguided.

Keywords: methylphenidate, stimulants, evidence accumulation, conflict tasks, executive functions, cognitive modeling, computational psychiatry, Bayesian

Methylphenidate and other stimulants are among the most effective pharmacological interventions for symptoms of attention-deficit/hyperactivity disorder (ADHD: Pliszka & AACAP Work Group on Quality Issues, 2007; Stuhec, Munda, Svab, & Locatelli, 2015). The pharmacological action of stimulants at the cellular level, which involves blocking dopamine and norepinephrine reuptake from the synapse, has been well-characterized for decades (Solanto, 1998). However, the precise mechanisms through which these biophysical processes alter complex behavioral outcomes are still only partially understood.

A wealth of previous research demonstrates that acute stimulant administration increases accuracy and response speed on choice reaction time tasks and improves performance across a variety of paradigms thought to index complex cognitive functions implicated in ADHD, including attention, working memory and response inhibition (e.g., Coghill et al., 2014; Pietrzak, Mollica, Maruff, & Snyder, 2006; Reid & Borkowski, 1984; Rosch et al., 2016; Strand et al., 2010). Recently, Hawk and colleagues (2018) provided the first direct empirical evidence that methylphenidate-related improvements in cognitive performance partially mediate drug effects on clinical outcome measures in ADHD. Although this line of work supports the hypothesis that stimulants’ effects on cognition are an important component of their clinical utility, an explanatory gap remains; a biologically-plausible account of how stimulant effects on catecholamine action improve performance across a variety of higher-order cognitive domains has yet to be elucidated.

One candidate account has been suggested by recent work within the framework of formal evidence accumulation models, which explain response time and accuracy on choice tasks as the product of a race between accumulators that gather noisy evidence for each possible choice over time until one choice “wins” by reaching a critical threshold of evidence (Smith & Ratcliff, 2004). Multiple models in this class have demonstrated considerable success in explaining patterns of behavioral data across a wide variety of cognitive tasks (e.g., Brown & Heathcote, 2008; Usher & McClelland, 2001; Ratcliff & McKoon, 2008). Their biological plausibility is strongly supported by single-cell recordings in primates (Gold & Shadlen, 2007; Hanes & Schall, 1996; Smith & Ratcliff, 2004), which suggest that spiking activity within certain neural subpopulations during choice tasks displays properties consistent with an evidence accumulation process.

Given the importance of evidence accumulation for cognition, it is possible that methylphenidate and other stimulants improve cognitive task performance by enhancing evidence accumulation through their influence on catecholamine systems that modulate neural signal-to-noise ratios. This idea found support in a recent study by Fosco, White and Hawk (2017) in a model-based analysis of cognitive performance in children with ADHD, a condition associated with less efficient evidence accumulation overall (Ziegler, Pedersen, Mowinckel, & Biele, 2016; Weigard, Huang-Pollock, Brown & Heathcote, 2018). They found that the drug dramatically increased children’s efficiency of evidence accumulation toward correct choices, reduced their threshold for responding (i.e., causing a faster, less cautious response style), and lengthened time spent on non-decision processes (e.g., motor speed). In addition, using electrophysiological methods, Loughnane et al. (2019) found that stimulant administration prior to oddball task performance in healthy adults reduces the latency and increases the build-up rate of the P3b potential, which is thought to correspond to evidence accumulation (Kelly & O’Connell, 2013).

These observations support the promising hypothesis that methylphenidate and other stimulants enhance evidence accumulation by selectively targeting relevant evidence (increasing signal) and/or by reducing variability in accumulation processes (decreasing noise). However, a key gap in existing work is that it has been limited to investigation of relatively simple perceptual tasks, for which most evidence accumulation models were designed. In contrast, most tasks in the literature on stimulants and ADHD involve more complex paradigms that are designed to explicitly measure specific top-down interference resolution processes thought to be aberrant in ADHD (Coghill et al., 2014; Willcutt et al., 2005). For example, in “conflict” tasks, where irrelevant stimulus attributes, such as location in the Simon task (Hedge & Marsh, 1975) or nearby visual stimuli in the Flanker task (Eriksen & Eriksen, 1974), conflict with relevant attributes, a top-down interference resolution process is theorized to overcome the influence of irrelevant information. Hence, an important question concerns the selectivity of the drug’s effects: does methylphenidate improve performance generally across both conflict and non-conflict tasks (or conflict and non-conflict conditions within a task), or are its effects at least partially selective to the top-down interference resolution process that is postulated to be operative under conditions of conflict? Evidence accumulation models could be highly useful in addressing this question, but extending these models to conflict tasks has proven challenging. There have been several important attempts to formulate evidence accumulation models for such tasks (Hübner, Steinhauser, & Lehle, 2010; Ulrich, Schröter, Leuthold, & Birngruber, 2015; White, Ratcliff & Starns, 2011), but their parameter recovery has thus far been found to be relatively poor (White, Servant & Logan, 2018), which significantly limits their use as measurement tools (Heathcote, Brown & Wagenmakers, 2015).

Recent work by Heathcote, Hannah and Matzke (under review) instead used existing parameters from the linear ballistic accumulator (LBA: Brown & Heathcote, 2008), a well-established evidence accumulation model, to explain behavioral effects on standard conflict tasks. The LBA (Figure 1a) frames decisions as a race between two or more accumulators (one for each possible choice) that gather evidence at a constant rate until one reaches a response threshold, denoted by a b parameter. On a given trial, the rate of accumulation for accumulators is drawn from a normal distribution with a mean of v and standard deviation of sv, which are both typically estimated separately for accumulators that match the stimulus (correct) and that mismatch the stimulus (incorrect). The model also contains parameters for uniformly-distributed start point noise (A) and time taken up by non-decision processes (t0).

Figure 1.

Figure 1

Open in a new tab

Schematic of the model. a) LBA model of a standard Simon task, where shape (square vs. triangle) is the relevant information, including the priming process proposed by the Conflict LBA: A = start point variability, b.pr = threshold for primed accumulator, b.non-pr = threshold for non-primed accumulator, v = mean drift rate, sv = drift rate variability, mat = accumulator matching the stimulus, mis = accumulator mismatching the stimulus. b) Conflict LBA model used to explain behavior on the MSIT, which includes a similar (Simon-like) priming effect to that displayed above and also allows drift rate parameters to vary by congruency condition. Note that, in our primary analyses, we only allowed v to vary by congruency condition in the MDR model and only allowed sv to vary by congruency condition in the DRV model. For drift rate parameter labels: c = congruent, i = incongruent.

Importantly, to account for conflict effects, Conflict LBA theory (Heathcote et al., under review) proposes that early interference from irrelevant stimulus attributes can be described as a priming effect that gives the accumulator for the choice favored by these attributes a head start in the race. Since increases in LBA start points are identical to threshold reductions, this effect is reflected by reductions in b for the matching (correct) accumulator on congruent trials and for the mismatching (error) accumulator on incongruent trials (Figure 1a). The theory further assumes that a top-down process attempts to correct for irrelevant information during the trial, but that this correction is often imperfect and varies substantially from trial to trial. This variability results in greater sv on incongruent trials while miscalibration in the top-down process is reflected by altered (typically reduced) v on incongruent trials. Heathcote et al. (under review) demonstrated that different combinations of these mechanisms (priming, plus calibration and variability in interference resolution processes) accounted for hallmark phenomena (e.g., delta function shapes: de Jong, Liang, & Lauber, 1994) across a variety of conflict tasks.

The current study applies the Conflict LBA to data from healthy adults who completed the Multi-Source Interference Task (MSIT: Bush & Shin, 2006) in a double-blind, placebo-controlled experimental trial of methylphenidate. We aimed to test whether methylphenidate effects on evidence accumulation seen in prior studies can be extended to conflict tasks. In addition, we were interested in whether methylphenidate affects the integrity of evidence accumulation in both congruent and incongruent task conditions or, alternatively, selectively facilitates processes in the incongruent condition, such as the top-down interference resolution process thought to operate in this condition.

Methods

Sample and Experimental Procedures

48 healthy participants were recruited from the community through ads on University of Michigan websites and flyers placed on campus and in other Ann Arbor locations. Of this initial sample, two participants were excluded for failing to complete one of the experimental sessions and an additional participant was excluded due to apparent misunderstanding of the task (<50% accuracy in the incongruent condition at one session), leaving 45 participants (age 21.9 +/− 3.6 years, range 18–33; 24 females) for analysis.

All experimental procedures were approved by a local institutional review board and were in accordance with the ethical standards of the Declaration of Helsinki. In a double-blind, randomized, cross-over design, participants received either 40mg of methylphenidate (MPD) or a placebo (PBO) 80 minutes prior to performing an event-related MSIT task during fMRI scanning. Herein, we focus exclusively on modeling the behavioral data because we were concerned that fMRI results would be beyond the scope of the study and make the current report overly complex. Order of drug administration (MPD vs. PBO) was counter-balanced across subjects. The dose of MPD was chosen to optimize predicted effects on task performance in an acute dosing context, consistent with recent studies (Clatworthy, et al. 2009, Schlösser, et al. 2009). In particular, dosing was adjusted upwards relative to clinical dosing schedules to account for the fact that clinical dosing takes advantage of chronic administration to achieve higher steady-state blood levels.

The MSIT was modeled after the version of the task used by Bush et al. (2008). Participants were presented with white three-digit stimuli on a black background and were instructed to press a response button (1, 2 or 3) to indicate which digit was unique (e.g., respond “2” for the stimulus 020). In the congruent condition, the unique digit was always in the same serial position as the correct response and the other digits were 0s (e.g., 100, 003), while in the incongruent condition the position of the unique digit was mismatched with the that of the correct response and other digits represented competing responses (e.g., 233, 212). Participants completed 100 trials in each congruency condition, which were pseudo-randomly interspersed, at each experimental session. Subjects were told to respond to stimuli rapidly and accurately. During each trial, stimuli appeared for 500ms, which was followed by a jittered ITI of 1–8.5s.

Specification of Models

The MSIT paradigm contains two possible forms of interference: a Simon-like effect, which primes the choice matching the serial position of the target in both congruent and incongruent conditions, and a Flanker-like effect, which produces interference in the incongruent condition only, as the flanking elements in the congruent condition (0s) do not correspond to a possible response. The Conflict LBA was previously found to describe behavioral data from both the Simon and Flanker paradigms well (Heathcote et al., under review), suggesting that this framework could jointly account for both types of effects on the MSIT. However, the LBA, like many cognitive and neuroscience models, displays highly-correlated parameters that can be difficult to estimate when individual-level data are sparse (Kolossa & Kopp, 2018; Gutenkunst et al., 2007). As a result, inclusion of all Conflict LBA parameters in situations where the number of trials in each condition is small and/or when errors are rare may result in parameter tradeoffs that compromise interpretability (e.g., Heathcote, Loft & Remington, 2015), making simplified models more useful (Heathcote et al., under review). A key concern for the current analysis is that sv parameters are particularly difficult to estimate in evidence accumulation models, in part because they trade off with v parameters, leading to inaccurate or biased estimates of both (Boehm et al., 2018).

As the MSIT had relatively few trials in each cell and low error rates, we sought to fit simplified versions of the Conflict LBA. On the basis of results from a targeted model selection analysis and several practical considerations, all of which are reported in detail in Supplemental Materials, we selected two models for use in the primary analyses: 1) a “mean drift rate” (MDR) model, in which v was allowed to vary by match/mismatch, congruency condition and drug condition, while sv was only allowed to vary by match/mismatch, and 2) a “drift rate variability” (DRV) model in which sv was allowed to vary by match/mismatch, congruency condition and drug condition, while v was only allowed to vary by match/mismatch. Results from both models are reported below because parameter tradeoffs would likely prevent effects in v and sv from being easily distinguished from one another in models of this data set. Hence, we were able to assess whether evidence accumulation processes were enhanced by the drug or were enhanced in the congruent, relative to the incongruent, condition, but we were limited in our ability to make specific inferences about whether these improvements were due to increases in average rates of evidence accumulation or reductions in the variability of evidence accumulation.

We fixed sv for the mismatching accumulator in the MDR model and sv for the congruent condition mismatching accumulators in the DRV model to 1 in order to identify parameters (see Donkin, Brown & Heathcote, 2009). Due to the previous findings of stimulant effects on response thresholds and non-decision time (Fosco et al., 2017), both models also allowed b and t0 to also vary by drug condition (MPD/PBO). Both models allowed b to explain priming from the Simon-like effect only (Figure 1b) due to practical challenges (Supplemental Materials) with allowing b to simultaneously explain the Flanker-like effect; nine b parameters were estimated in each drug condition, with different thresholds for each accumulator when the target was located at each of the three possible positions. This allowed for both priming (i.e., lower b for the accumulator corresponding to the target position) as well as response bias (i.e., different thresholds for each accumulator, allowing a lower b for those corresponding to favored responses). A single A parameter was estimated for each model across all conditions.

Parameter Estimation and Hypothesis Testing

Prior to model-fitting, RTs <300ms were removed as likely fast guesses (<1% of data). Due the task’s low error rate, we used hierarchical Bayesian methods in Dynamic Models of Choice (DMC: Heathcote et al., 2018) to fit the Conflict LBA, as such methods allow estimation of group-level parameters that can be used for inference and inform individual-level estimates, even when parameters are not well-identified at the individual level alone. In this model, group distributions of all LBA parameter values were assumed to follow truncated positive normal distributions described by location (μ) and scale (σ) hyper-parameters, for which broad and non-informative priors were posited (all priors reported in Supplemental Materials). Differential evolution Markov chain Monte Carlo simulations (Turner, Sederberg, Brown & Steyvers, 2013), with start points determined by earlier simulations at the individual level, were used to sample from posterior distributions of all group- and individual-level parameter values. Sixty-five separate chains were used in the simulations, and thinning (retaining only every 10th sample) was implemented to save file space. After an initial burn-in period that lasted until convergence was indicated by both visual inspection of chains (to ensure they were overlapping and stable) and the Gelman-Rubin statistic (G-R<1.10: Gelman & Rubin, 1992), 300 iterations of the simulation were retained, leaving 19,500 posterior samples per-parameter for inference.

Model fit was assessed with posterior predictive plots (Gelman, Meng & Stern, 1996), which allow visual inspection of how well the model accounts for key effects in behavioral data. Effects in parameter values were assessed using 95% credible intervals (CIs) of average posterior difference distributions of individual-level parameters. We considered effects to be reliable if the 95% CI did not contain 0. Procedures for calculating posterior difference distributions are described in Supplemental Materials. All code and data are available at: osf.io/t3cn5.

Results

Model Fit

Analyses of behavioral summary statistics (Supplemental Materials) indicated that RTs were longer and accuracy was lower in the incongruent, relative to congruent conditions, and that RTs were shorter in the MPD, relative to PBO, conditions. Plots of posterior predictive data compared with empirical data for group average correct1 RT quantiles (.1, .5 and 9) and group average accuracy rates are displayed in Figure 2 for both models. Black points and lines represent empirical data while gray violin density plots represent the range of values predicted by 500 samples drawn from posterior distributions of model parameters. Therefore, the density and spread of the violin plots represent uncertainty in predictions of the model due to uncertainty about the values of its parameter estimates. Both models provide an excellent description of correct RT quantiles and capture the RT main effects present in empirical data: shorter RT in the MPD condition and longer RT in the incongruent condition. Although both models captured the general pattern of lower accuracy rates in the incongruent condition well, predicted absolute values of accuracy rates in this condition were lower than those in the empirical data. However, the difference was of a small magnitude (2–3%), suggesting that this misfit is relatively minor.

Figure 2.

Figure 2

Open in a new tab

Posterior predictive plots of group average correct RT quantile and accuracy rate summary statistics from the empirical data (black points and lines) plotted with those predicted by 500 posterior samples (represented by gray violin density plots) from the MDR (left) and DRV (right) models.

Figure 3 displays group-averaged delta functions, which plot the difference between congruent and incongruent correct RT quantiles (.1, .3, .5, .7 and .9) on the y-axis against their mean on the x-axis, and conditional accuracy functions (CAFs), which display mean accuracy rates as a function of RT quantile bins. Delta functions showed a linear increasing pattern in both drug conditions, which is atypical for most variants of the Simon task and more common in Flanker paradigms (Heathcote et al., under review). Nonetheless, both models captured this pattern well, suggesting that they provided an excellent description of interference effects in RT. CAFs of the PBO condition indicated that fast (<.25 quantile) and slow (>.75 quantile) incongruent condition responses were slightly less accurate than responses in the middle of the range. CAFs of the MPD condition showed a clearer trend of reduced accuracy for fast responses, suggesting fast errors due to a possible speed/accuracy trade-off in this condition. The DRV model appeared to provide a better description of these CAF trends than the MDR model.

Figure 3.

Figure 3

Open in a new tab

Group average delta functions and conditional accuracy functions (CAFs) for the empirical data and for values predicted by the MDR (left) and DRV (right) models. Delta functions represent the difference between congruent and incongruent correct RT quantiles (.1, .3, .5, .7 and .9) on the y-axis against the congruent/incongruent mean on the x-axis, for empirical data (black points and lines) and data predicted by 500 posterior samples from the model (small gray points = data predicted by each individual sample, gray lines = average of data predicted by all posterior samples). CAFs display accuracy rates for different RT quantile bins in the congruent (dotted lines) and incongruent (solid lines) conditions. Black lines and points represent accuracy rates for each bin in the empirical data while gray violin plots represent accuracy rates predicted by 500 posterior samples from the model. For all plots: Con = congruent, Inc = incongruent, MPD = methylphenidate condition, PBO = placebo condition.

Model Parameter Estimates

We focused on drug and congruency condition effects in several main parameters of interest, which are displayed in Figures 4 and 5 for the MDR and DRV models, respectively. For the MDR model, we assessed effects in both the mean quantity of evidence individuals are able to accumulate (the average of v.match and v.mismatch) and the mean quality of that evidence (v.match minus v.mismatch). Increases in evidence quantity mostly reduce response times, while increases in evidence quality mostly lead to more accurate responding. For the DRV model, we separately assessed effects in drift variability for the matching (sv.match) and mismatching (sv.mismatch) accumulators. For both MDR and DRV models, drift rate effects were also assessed using a sensitivity metric (Heathcote el al., 2015; Winkel et al., 2016), which is similar to d’ from signal detection theory in that it indexes the difference between mean rates of the matching and mismatching accumulators (signal) relative to their variability (noise):

sensitivity=(v.match−v.mismatch)/sv.match2+sv.mismatch2/2

This summary metric provides an index of discrimination between choices in an accumulator model that can be compared across the MDR and DRV models despite the fact that, as noted above, congruency or drug effects in sensitivity cannot be specifically attributed to either v or sv differences given limitations of the current data set. For ease of interpretation, we averaged b parameter estimates for accumulators that were primed and separately averaged those that were not primed, with the goal of investigating effects of priming and drug condition on individuals’ response thresholds in both MDR and DRV models. Finally, we assessed drug effects on non-decision time (t0) in both MDR and DRV models. For inference, medians (Δ) and 95% CIs of posterior difference distributions for all effects are reported below. Posterior difference distributions of interactions indicate the main effects of priming (primed – non-primed) or congruency condition (congruent - incongruent) in the PBO condition subtracted from the same effects in the MPD condition.

Figure 4.

Figure 4

Open in a new tab

Medians (gray bars) and 95% credible intervals (black error bars) for average posterior distributions of MDR model parameters of interest in the placebo (PBO) and methylphenidate (MPD) conditions. Parameters including overall evidence quantity (average of v.match and v.mismatch), the congruency effect in evidence quantity, overall evidence quality (v.match minus v.mismatch), the congruency effect in evidence quality, overall perceptual sensitivity, the congruency effect in perceptual sensitivity, overall response threshold (b), priming effects in response threshold, and non-decision time (t0). CIs were obtained by averaging over samples from posterior distributions of individual-level parameter values.

Figure 5.

Figure 5

Open in a new tab

Medians (gray bars) and 95% credible intervals (black error bars) for average posterior distributions of DRV model parameters of interest in the placebo (PBO) and methylphenidate (MPD) conditions. Parameters including overall rate variability of the matching accumulator (sv.match), the congruency effect in sv.match, overall rate variability of the mismatching accumulator (sv.mismatch), the congruency effect in sv.mismatch, overall perceptual sensitivity, the congruency effect in perceptual sensitivity, overall response threshold (b), priming effects in response threshold, and non-decision time (t0). CIs were obtained by averaging over samples from posterior distributions of individual-level parameter values.

MDR model.

Evidence quantity was greater overall in the congruent condition, Δ =.36, CI = [.32, .41], and greater in the MPD condition, Δ =.19, CI = [.14, .24]. There was also evidence for a congruency x drug interaction, Δ =.14, CI = [.05, .23], in which the congruency-related differences in evidence quantity were of greater magnitude in the MPD condition, Δ =.43, CI = [.36, .50], than in PBO, Δ =.29, CI = [.24, .35]. Similarly, evidence quality was greater overall in the congruent condition, Δ =.41, CI = [.32, .51] and greater overall in the MPD condition, Δ =.13, CI = [.04, .22]. A congruency x drug interaction in which congruency-related effects on evidence quality were lower in MPD than in PBO was not reliably different from 0, Δ =−.15, CI = [−.33, .03]. Sensitivity was greater overall in the congruent condition, Δ =.47, CI = [.36, .57], and greater in the MPD condition, Δ =.15, CI = [.04, .25], likely because of the increased evidence quality present in both conditions. A congruency x drug interaction effect in sensitivity was not reliably different from 0, Δ =−.17, CI = [−.37, .03]. Response thresholds were lower for primed accumulators, Δ =−.63, CI = [−65, −.61], and lower overall in the MPD condition, Δ =−.18, CI = [−20, −.16]. A priming x drug interaction, Δ =−.06, CI = [−.10, −.02], indicated that the priming effect was larger in the MPD, Δ =−.66, CI = [−.69, −.64], than PBO condition, Δ =−.60, CI = [−.63, −.58]. Non-decision times were longer in MPD, Δ =.037 seconds, CI = [.030, .043].

DRV model.

For sv.match, rates were less variable in the congruent condition, Δ = −.05, CI = [−.06, −.03], and less variable in the MPD condition, Δ = −.07, CI = [−.09, −.06]. There was evidence for a congruency x drug interaction, Δ = −.04, CI = [−.07, −.02], in which worsening of sv.match in the incongruent condition was of greater magnitude in the MPD condition, Δ = −.07, CI = [−.09, −.05], than in the PBO condition, Δ = −.03, CI = [−.04, −.01]. For sv.mismatch, rates were more variable in the congruent condition, Δ = .27, CI = [.25, .29], while the drug effect was not credibly different from 0, Δ = −.02, CI = [−.04, .00]. As sv.mismatch was fixed to 1 in the congruent condition, interactions in this parameter were not assessed. Sensitivity was greater overall in the MPD condition, Δ =.28, CI = [.18, .38], consistent with findings from the MDR model, although increased sensitivity in the DRV model was due to reduced drift rate variability rather than to improved mean quality of evidence. In contrast to MDR model findings, sensitivity was lower in the congruent condition, Δ = −.88, CI = [−.99, −.78], and there was evidence for a congruency x drug interaction, Δ = −.28, CI = [−.48, −.08] in which congruency-related reductions in sensitivity were greater for the MPD, Δ =−1.02, CI = [−1.20, −.86], than for the PBO condition, Δ = −.74, CI = [−.87, −.62]. However, this interaction should be interpreted with caution due to the paradoxical direction of the congruency main effect (which implies better performance in the incongruent condition), and the fact that sv.mismatch was fixed to 1 across congruent trials, which may have led to reduced estimates of MPD-related sensitivity effects on these trials. Thresholds were lower for primed accumulators, Δ =−.70, CI = [−71, −.69], and lower in the MPD condition, Δ =−.14, CI = [−15, −.12]. A priming x drug interaction effect was not credibly different from 0, Δ =.00, CI = [−.02, .02]. The drug effect on non-decision time was not credibly different from 0, Δ =.001 seconds, CI = [−.006, .007].

Discussion

The current study examined MSIT task data from healthy adults who received acute methylphenidate challenge in a randomized, placebo-controlled, double-blind, cross-over study. We applied two variants of the Conflict LBA (Heathcote et al., under review), a framework that extends the established LBA model (Brown & Heathcote, 2008) to describe interference-related processes in conflict paradigms. Consistent with previous research on simple perceptual tasks (Fosco et al., 2017; Loughnane et al., 2019), we found that methylphenidate improved performance by reducing response thresholds and by enhancing the sensitivity (signal-to-noise ratio) of evidence accumulation processes, although we were unable to pinpoint whether evidence accumulation was improved due to increases in the average quality of evidence (signal) or to reductions in evidence accumulation variability (noise). Our results suggest that enhancement of evidence accumulation, in combination with associated reductions in response threshold, is a biologically-plausible, task-general mechanism through which stimulants improve cognition.

In contrast, methylphenidate did not appear to selectively influence processes that were directly probed by the experimental conflict manipulation. Differences between congruent and incongruent trials on the MSIT were accounted for both by a priming effect in thresholds, which biased early evidence toward choices indicated by irrelevant information, and by differences in drift rate processes between congruent and incongruent trials. Alternate models suggested that either 1) evidence quality and quantity were reduced in the incongruent condition (MDR model) or 2) rates for the matching accumulator become more variable, and those for the mismatching accumulator became less variable, in the incongruent condition (DRV model). The former effects may reflect miscalibration or insufficiency of interference resolution processes in the incongruent condition, while the latter may reflect variability in these processes (Heathcote et al., under review). Although inferences varied somewhat between the two models, there was agreement across models that methylphenidate did not reduce either the priming effect caused by irrelevant information or differences in evidence accumulation between congruent and incongruent trials. Indeed, when drug-related interaction effects were found, there was evidence that priming effects and congruency-related differences in drift rate variables were actually increased by the drug.

Notably, the presence of fast errors and reductions in response thresholds under stimulant challenge suggests that participants implemented speed/accuracy trade-offs in conditions where their evidence accumulation was improved, which are considered adaptive because they often further reduce RT without a significant decrement to accuracy rates (Dutilh, Vandekerckhove, Tuerlinckx, & Wagenmakers, 2009). However, it is unclear whether such adaptive threshold reductions are a direct effect of the drug, or simply a secondary response to drug-related enhancements in evidence accumulation. It is also notable that, despite the general consistency of our findings with those of Fosco et al. (2017), only the MDR, and not the DRV, model replicated their finding of increased non-decision times under stimulant challenge. The fact that this effect, which is difficult to interpret (Fosco et al., 2017) and mirrors paradoxical findings of shorter non-decision times in ADHD (Karalunas, Geurts, Konrad, Bender, & Nigg, 2014; Weigard et al., 2018), was not robust across models suggests that it may result from parameter trade-offs, although additional work is needed to confirm this possibility.

Taken together, this pattern of results suggests that methylphenidate facilitates conflict task performance by enhancing evidence accumulation processes across all task conditions, rather than by selectively influencing specific processes, such as top-down interference resolution processes, that are proposed to be specifically activated in incongruent conditions. Such an account explains why methylphenidate-related facilitation effects are present across a wide variety of cognitive paradigms (Coghill et al., 2014; Pietrzak et al., 2006), including simple choice tasks (Reid & Borkowski, 1984). Moreover, it is consistent with an emerging set of findings from tasks that do not have overt manipulations aimed at taxing specific top-down processes (Fosco et al., 2017, Loughnane et al., 2019). When considered within the emerging body of work on methylphenidate and evidence accumulation processes, the current findings also support a biologically-plausible account of stimulant effects on behavior. Stimulants may increase the overall integrity of evidence accumulation across a wide variety of tasks and experimental conditions by facilitating catecholamine systems that optimize neural signal-to-noise ratios through modulation of arousal and/or neural gain, such as the locus coeruleus norepinephrine (LC-NE) system (Aston-Jones and Cohen, 2005). Such an account would be consistent with theories that implicate poor signal-to-noise ratios across a wide variety of tasks in ADHD due to deficits in catecholamine-mediated state-regulation processes (Karalunas et al., 2014; Weigard et al., 2018; Sikström & Söderlund, 2007). It would also be consistent with related theories that implicate metabolic limits on evidence accumulation efficiency in ADHD, which are hypothesized to be related to norepinephrine action (Killeen, Russell, & Sergeant, 2013).

Findings from this study should be considered within the context of several limitations. First, the analysis involved simplified versions of the Conflict LBA that assumed priming could account for interference caused by the Simon-like manipulation, only, and that were not able to distinguish effects in mean drift rates from those in drift rate variability. Although these choices were informed by model selection analyses and practical constraints that are detailed in Supplemental Materials, it is unclear whether they affected our substantive inferences. Second, our MSIT task had a very low error rate, which both made parameter estimation difficult and may prevent our findings from generalizing to tasks with higher error rates. Both limitations underscore the need to experimentally validate a measurement model of the MSIT task, which would likely involve changes to the task that both increase the number of trials per cell and generate errors for the purpose of improving parameter estimation. Finally, although this study focuses on modeling behavioral data, our key conclusions should also be tested at the neural level of analysis. Specifically, future work should determine whether methylphenidate’s effects on neuroimaging indices are also similar between conflict and non-conflict conditions.

In sum, we applied the Conflict LBA (Heathcote et al., under review), a novel mathematical model of conflict task performance, to MSIT data from healthy adults receiving acute methylphenidate or placebo challenge. We found stimulants improve task performance by enhancing evidence accumulation processes across both congruent and incongruent conditions, suggesting absence of selective facilitation of processes specific to the incongruent condition (e.g., top-down interference resolution processes). Our findings help clarify methylphenidate’s mechanism of action and invite further computational and neurobiological investigation.

Supplementary Material

213_2019_5316_MOESM1_ESM

Footnotes

Publisher's Disclaimer: This Author Accepted Manuscript is a PDF file of a an unedited peer-reviewed manuscript that has been accepted for publication but has not been copyedited or corrected. The official version of record that is published in the journal is kept up to date and so may therefore differ from this version.

Conflict of Interest Statement

On behalf of all authors, the corresponding author states that there is no conflict of interest.

1

Predictive plots of error RTs would be difficult to interpret due to low error rates, and are therefore not reported.

References

  1. Aston-Jones G, & Cohen JD (2005). An integrative theory of locus coeruleus-norepinephrine function: adaptive gain and optimal performance. Annu. Rev. Neurosci, 28, 403–450. [DOI] [PubMed] [Google Scholar]
  2. Barateau L, Lopez R, & Dauvilliers Y (2016). Management of narcolepsy. Current treatment options in neurology, 18(10), 43. [DOI] [PubMed] [Google Scholar]
  3. Boehm U, Annis J, Frank MJ, Hawkins GE, Heathcote A, Kellen D, … & van Ravenzwaaij D (2018). Estimating across-trial variability parameters of the Diffusion Decision Model: Expert advice and recommendations. Journal of Mathematical Psychology, 87, 46–75. [Google Scholar]
  4. Brown SD, & Heathcote A (2008). The simplest complete model of choice response time: Linear ballistic accumulation. Cognitive psychology, 57(3), 153–178. [DOI] [PubMed] [Google Scholar]
  5. Bush G, & Shin LM (2006). The Multi-Source Interference Task: an fMRI task that reliably activates the cingulo-frontal-parietal cognitive/attention network. Nature protocols, 1(1), 308. [DOI] [PubMed] [Google Scholar]
  6. Bush G, Spencer TJ, Holmes J, Shin LM, Valera EM, Seidman LJ, … & Biederman J (2008). Functional magnetic resonance imaging of methylphenidate and placebo in attention-deficit/hyperactivity disorder during the multi-source interference task. Archives of General Psychiatry, 65(1), 102–114. [DOI] [PubMed] [Google Scholar]
  7. Clatworthy PL, Lewis SJ, Brichard L, Hong YT, Izquierdo D, Clark L, … & Robbins TW (2009). Dopamine release in dissociable striatal subregions predicts the different effects of oral methylphenidate on reversal learning and spatial working memory. Journal of Neuroscience, 29(15), 4690–4696. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Coghill DR, Seth S, Pedroso S, Usala T, Currie J, & Gagliano A (2014). Effects of methylphenidate on cognitive functions in children and adolescents with attention-deficit/hyperactivity disorder: evidence from a systematic review and a meta-analysis. Biological psychiatry, 76(8), 603–615. [DOI] [PubMed] [Google Scholar]
  9. de Jong R, Liang CC, & Lauber E (1994). Conditional and unconditional automaticity: a dual-process model of effects of spatial stimulus-response correspondence. Journal of Experimental Psychology: Human Perception and Performance, 20(4), 731. [DOI] [PubMed] [Google Scholar]
  10. Donkin C, Brown SD, & Heathcote A (2009). The overconstraint of response time models: Rethinking the scaling problem. Psychonomic Bulletin & Review, 16(6), 1129–1135. [DOI] [PubMed] [Google Scholar]
  11. Dutilh G, Vandekerckhove J, Tuerlinckx F, & Wagenmakers EJ (2009). A diffusion model decomposition of the practice effect. Psychonomic Bulletin & Review, 16(6), 1026–1036. [DOI] [PubMed] [Google Scholar]
  12. Eriksen BA, & Eriksen CW (1974). Effects of noise letters upon the identification of a target letter in a nonsearch task. Perception & psychophysics, 16(1), 143–149. [Google Scholar]
  13. Fosco WD, White CN, & Hawk LW (2017). Acute Stimulant Treatment and Reinforcement Increase the Speed of Information Accumulation in Children with ADHD. Journal of abnormal child psychology, 45(5), 911–920. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Gelman A, Meng XL, & Stern H (1996). Posterior predictive assessment of model fitness vi a realized discrepancies. Statistica sinica, 733–760. [Google Scholar]
  15. Gelman A, & Rubin DB (1992). Inference from iterative simulation using multiple sequences. Statistical science, 7(4), 457–472. [Google Scholar]
  16. Gold JI, & Shadlen MN (2007). The neural basis of decision making. Annu. Rev. Neurosci, 30, 535–574. [DOI] [PubMed] [Google Scholar]
  17. Gutenkunst RN, Waterfall JJ, Casey FP, Brown KS, Myers CR, & Sethna JP (2007). Universally sloppy parameter sensitivities in systems biology models. PLoS computational biology, 3(10), e189. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Hanes DP, & Schall JD (1996). Neural control of voluntary movement initiation. Science, 274(5286), 427–430. [DOI] [PubMed] [Google Scholar]
  19. Hardy SE (2009). Methylphenidate for the treatment of depressive symptoms, including fatigue and apathy, in medically ill older adults and terminally ill adults. The American journal of geriatric pharmacotherapy, 7(1), 34–59. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Hawk LW Jr, Fosco WD, Colder CR, Waxmonsky JG, Pelham WE Jr, & Rosch KS (2018). How do stimulant treatments for ADHD work? Evidence for mediation by improved cognition. Journal of Child Psychology and Psychiatry [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Heathcote A, Brown SD, & Wagenmakers EJ (2015). An introduction to good practices in cognitive modeling. In An introduction to model-based cognitive neuroscience (pp. 25–48). Springer, New York, NY. [Google Scholar]
  22. Heathcote A, Hannah K, & Matzke D (under review). Priming and variable control in choice conflict tasks [Google Scholar]
  23. Heathcote A, Lin YS, Reynolds A, Strickland L, Gretton M, & Matzke D (2018). Dynamic models of choice. Behavior research methods, 1–25. [DOI] [PubMed] [Google Scholar]
  24. Heathcote A, Loft S, & Remington RW (2015). Slow down and remember to remember! A delay theory of prospective memory costs. Psychological Review, 122, 367–410. [DOI] [PubMed] [Google Scholar]
  25. Hedge A, & Marsh NWA (1975). The effect of irrelevant spatial correspondences on two-choice response-time. Acta psychologica, 39(6), 427–439. [DOI] [PubMed] [Google Scholar]
  26. Heathcote A, Suraev A, Curley S, Gong Q, Love J, & Michie PT (2015). Decision processes and the slowing of simple choices in schizophrenia. Journal of Abnormal Psychology, 124(4), 961. [DOI] [PubMed] [Google Scholar]
  27. Hübner R, Steinhauser M, & Lehle C (2010). A dual-stage two-phase model of selective attention. Psychological review, 117(3), 759. [DOI] [PubMed] [Google Scholar]
  28. Karalunas SL, Geurts HM, Konrad K, Bender S, & Nigg JT (2014). Annual research review: Reaction time variability in ADHD and autism spectrum disorders: Measurement and mechanisms of a proposed trans‐ diagnostic phenotype. Journal of Child Psychology and Psychiatry, 55(6), 685–710. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Kelly SP, & O’Connell RG (2013). Internal and external influences on the rate of sensory evidence accumulation in the human brain. Journal of Neuroscience, 33(50), 19434–19441. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Killeen PR, Russell VA, & Sergeant JA (2013). A behavioral neuroenergetics theory of ADHD. Neuroscience & Biobehavioral Reviews, 37(4), 625–657. [DOI] [PubMed] [Google Scholar]
  31. Kolossa A, & Kopp B (2018). Data quality over data quantity in computational cognitive neuroscience. NeuroImage, 172, 775–785. [DOI] [PubMed] [Google Scholar]
  32. Loughnane GM, Brosnan MB, Barnes JJ, Dean A, Nandam SL, O’Connell RG, & Bellgrove MA (2019). Catecholamine Modulation of Evidence Accumulation during Perceptual Decision Formation: A Randomized Trial. Journal of cognitive neuroscience, 1–10. [DOI] [PubMed] [Google Scholar]
  33. Matzke D, Hughes M, Badcock JC, Michie P, & Heathcote A (2017). Failures of cognitive control or attention? The case of stop-signal deficits in schizophrenia. Attention, Perception, & Psychophysics, 79(4), 1078–1086. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Pietrzak RH, Mollica CM, Maruff P, & Snyder PJ (2006). Cognitive effects of immediate-release methylphenidate in children with attention-deficit/hyperactivity disorder. Neuroscience & Biobehavioral Reviews, 30(8), 1225–1245. [DOI] [PubMed] [Google Scholar]
  35. Pliszka S, & AACAP Work Group on Quality Issues. (2007). Practice parameter for the assessment and treatment of children and adolescents with attention-deficit/hyperactivity disorder. Journal of the American Academy of Child & Adolescent Psychiatry, 46(7), 894–921. [DOI] [PubMed] [Google Scholar]
  36. Ratcliff R, & McKoon G (2008). The diffusion decision model: theory and data for two-choice decision tasks. Neural computation, 20(4), 873–922. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Ratcliff R, Smith PL, Brown SD, & McKoon G (2016). Diffusion decision model: Current issues and history. Trends in cognitive sciences, 20(4), 260–281. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Reid MK, & Borkowski JG (1984). Effects of methylphenidate (Ritalin) on information processing in hyperactive children. Journal of abnormal child psychology, 12(1), 169–185. [DOI] [PubMed] [Google Scholar]
  39. Rosch KS, Fosco WD, Pelham WE, Waxmonsky JG, Bubnik MG, & Hawk LW (2016). Reinforcement and stimulant medication ameliorate deficient response inhibition in children with attention-deficit/hyperactivity disorder. Journal of abnormal child psychology, 44(2), 309–321. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Schlösser RGM, Nenadic I, Wagner G, Zysset S, Koch K, & Sauer H (2009). Dopaminergic modulation of brain systems subserving decision making under uncertainty: a study with fMRI and methylphenidate challenge. Synapse, 63(5), 429–442. [DOI] [PubMed] [Google Scholar]
  41. Sikström S, & Söderlund G (2007). Stimulus-dependent dopamine release in attention-deficit/hyperactivity disorder. Psychological review, 114(4), 1047. [DOI] [PubMed] [Google Scholar]
  42. Smith PL, & Ratcliff R (2004). Psychology and neurobiology of simple decisions. Trends in neurosciences, 27(3), 161–168. [DOI] [PubMed] [Google Scholar]
  43. Solanto MV (1998). Neuropsychopharmacological mechanisms of stimulant drug action in attention-deficit hyperactivity disorder: a review and integration. Behavioural brain research, 94(1), 127–152. [DOI] [PubMed] [Google Scholar]
  44. Stuhec M, Munda B, Svab V, & Locatelli I (2015). Comparative efficacy and acceptability of atomoxetine, lisdexamfetamine, bupropion and methylphenidate in treatment of attention deficit hyperactivity disorder in children and adolescents: a meta-analysis with focus on bupropion. Journal of affective disorders, 178, 149–159. [DOI] [PubMed] [Google Scholar]
  45. Strand MT, Hawk LW, Bubnik M, Shiels K, Pelham WE, & Waxmonsky JG (2012). Improving working memory in children with attention-deficit/hyperactivity disorder: the separate and combined effects of incentives and stimulant medication. Journal of abnormal child psychology, 40(7), 1193–1207. [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Turner BM, Sederberg PB, Brown SD, & Steyvers M (2013). A method for efficiently sampling from distributions with correlated dimensions. Psychological methods, 18(3), 368. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Ulrich R, Schröter H, Leuthold H, & Birngruber T (2015). Automatic and controlled stimulus processing in conflict tasks: Superimposed diffusion processes and delta functions. Cognitive Psychology, 78, 148–174. [DOI] [PubMed] [Google Scholar]
  48. Usher M, & McClelland JL (2001). The time course of perceptual choice: the leaky, competing accumulator model. Psychological review, 108(3), 550. [DOI] [PubMed] [Google Scholar]
  49. Weigard A, Huang-Pollock C, Brown S, & Heathcote A (2018). Testing formal predictions of neuroscientific theories of ADHD with a cognitive model–based approach. Journal of abnormal psychology, 127(5), 529. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. White CN, Ratcliff R, & Starns JJ (2011). Diffusion models of the flanker task: Discrete versus gradual attentional selection. Cognitive psychology, 63(4), 210–238. [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. White CN, Servant M, & Logan GD (2018). Testing the validity of conflict drift-diffusion models for use in estimating cognitive processes: A parameter-recovery study. Psychonomic bulletin & review, 25(1), 286–301. [DOI] [PMC free article] [PubMed] [Google Scholar]
  52. Willcutt EG, Doyle AE, Nigg JT, Faraone SV, & Pennington BF (2005). Validity of the executive function theory of attention-deficit/hyperactivity disorder: a meta-analytic review. Biological psychiatry, 57(11), 1336–1346. [DOI] [PubMed] [Google Scholar]
  53. Winkel J, Hawkins GE, Ivry RB, Brown SD, Cools R, & Forstmann BU (2016). Focal striatum lesions impair cautiousness in humans. Cortex, 85, 37–45. [DOI] [PubMed] [Google Scholar]
  54. Ziegler S, Pedersen ML, Mowinckel AM, & Biele G (2016). Modelling ADHD: A review of ADHD theories through their predictions for computational models of decision-making and reinforcement learning. Neuroscience & Biobehavioral Reviews, 71, 633–656. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

213_2019_5316_MOESM1_ESM
NIHMS1534632-supplement-213_2019_5316_MOESM1_ESM.docx (46.7KB, docx)

RESOURCES