Is ADHD a Valid Entity?

Redefining a Common Diagnosis Through Classical Validators and Computational Phenotyping

ADHD is everywhere. It’s one of the most common psychiatric diagnoses in both kids and adults, with prevalence estimates around 5–7% in youth and lower in adults (Willcutt, 2012). Stimulant medications reliably ease symptoms, and the diagnosis shows up in millions of charts. On the surface, that makes ADHD look like a well-established condition. But walk into any ADHD clinic and you’ll notice that many patients look nothing alike.

A seven-year-old boy can’t sit still in class. He blurts out answers, forgets his homework, and frustrates his teacher enough that she pushes for an evaluation. He meets criteria for ADHD, starts methylphenidate, and the change is immediate: his grades pick up, his teacher notices better behavior, and his parents feel relief. Fast forward a few years, by fourteen he’s off medication and doing fine academically and socially. So what exactly happened? Did he have ADHD that went into remission, or was it something else that only looked like ADHD for a few years?

Now consider a thirty-two-year-old woman who comes in for lifelong trouble with focus, procrastination, and keeping her life organized. She remembers being “spacey” as a child, but no one ever thought to have her evaluated. On paper, she meets DSM criteria for adult ADHD, and stimulants help her concentration. But there’s a catch: her sister has bipolar disorder, and she herself has had episodes of depression. When you check her childhood records, there’s no evidence of hyperactivity or impulsivity. So is this really “adult-onset ADHD,” or is her inattention just one face of mood instability?

Or take a twelve-year-old boy with serious behavioral problems: he can’t control his impulses, can’t focus on schoolwork, and clearly meets ADHD criteria. But he’s also anxious, very anxious, and stimulants only make it worse. Behavioral strategies help a little, and his symptoms rise and fall with the level of stress at home. Is ADHD the core problem here, or is his inattention secondary to anxiety?

Factor-analytic work consistently recovers an attention–hyperactivity dimension from symptom ratings, and medications produce acute benefit. But do these three patients have the same underlying condition? To answer that question, psychiatry has a set of tools inherited from general medicine.

What the Classical Validators Show

Since Robins and Guze (1970), psychiatry has tried to use validators beyond symptoms to test whether diagnostic categories represent valid entities. The idea came from general medicine: symptoms like fever or cough point toward several diseases, each requiring independent validators.

Let’s give a few examples so this is clear. Before bacteriology, “fevers” were treated as an entity. Typhoid, typhus, malaria, and tuberculosis all presented with high fever, but only became obviously distinct when we were able to culture Salmonella, see typhoid’s intestinal perforations at autopsy, identify malaria parasites in blood or notice its periodic pattern of relapse. Critically, they responded to different treatments: quinine for malaria, antibiotics for typhoid and tuberculosis.

A same thing happened with “dropsy,” the old term for swelling and fluid retention. Patients with edema were once grouped together, until validators showed that heart failure produced progressive shortness of breath and responded to digitalis, kidney disease caused proteinuria and carried a different prognosis, and cirrhosis of the liver produced ascites with varices and poor outcomes (apologies for the medical terms, I hope the ideas are clear). Each followed its own course and demanded different management.

“Chlorosis” or “green sickness” offers another example. It referred to pallor and fatigue in young women. Over time it gave rise to iron deficiency anemia, pernicious anemia with B12 deficiency and antibodies, sickle cell disease with genetic inheritance and vaso-occlusive crises, and others. Each showed specific markers, distinct complications, and distinct treatments.

The lesson is that a disease is validated when multiple lines of evidence (course of illness, genetics, biological markers, treatment response) converge on a coherent entity with predictable patterns. The question for ADHD is whether applying these validators shows us one disease, several distinct ones, or something that resists the disease framework altogether.

Start with symptom specificity. Inattention and executive dysfunction are not unique to ADHD. They show up in anxiety, depression, mania, and psychosis. In the National Comorbidity Survey Replication, most adults with ADHD had at least one additional disorder, with anxiety and mood disorders especially common (Kessler et al., 2006). Factor analyses do pull out a reproducible attention–hyperactivity dimension, but the problem is that the same dimension cuts across other disorders too.

Genetics tells a similar story. Twin studies show high heritability, but genome-wide association studies reveal that ADHD is highly polygenic, with small effect sizes scattered across the genome. Just as importantly, those genetic signals overlap heavily with depression, autism, and other conditions.

The course of illness is also heterogeneous. Most children diagnosed with ADHD do not continue to meet criteria as adults. In the Pelotas birth cohort, for example, only 17% of children with ADHD still qualified as young adults, and only 13% of adults with ADHD had the disorder in childhood (Caye et al., 2016). Some children clearly persist, but others remit, and many adults who meet criteria never showed it in childhood. This pattern suggests we are not tracking a single disease trajectory.

Biological markers tell a similar story. Structural neuroimaging shows small group-level differences, but with huge overlap between cases and controls. Shaw and colleagues (2007) found that ADHD was best characterized by delayed cortical maturation, especially in prefrontal regions. Crucially, many children caught up by adolescence. This looks less like a fixed structural abnormality and more like a developmental variant that often normalizes.

Treatment response is reliable but nonspecific. Stimulants work, but not the way antibiotics cure an infection. They improve attention in healthy controls, in people with depression, in people with schizophrenia. The MTA trial showed strong early benefits that faded over time. By six to eight years, there were no differences between treatment groups on any clinically relevant outcomes (Molina et al., 2009). The drugs help acutely, but they do not change the long-term trajectory.

In sum, the validators give us a mixed message. They show that the ADHD label does pick out a reproducible symptom cluster, but that it is broad, overlapping, and heterogeneous. It cannot be one disease in the classical medical understanding. It looks more like a set of computational problems that keep showing up across people in somewhat different forms. Framed that way, perhaps we have a better path forward?

From Symptoms to Algorithms

Let’s go back to the three patients. Standard diagnostic practice asks: Do they meet symptom criteria? Do symptoms cause impairment? Did symptoms start before age twelve? But what if we could measure something more fundamental about how their minds are working?

Let’s consider what may be giving rise to dysfunction (which is what we ultimately care about in psychiatry— processing information differently in and of itself is not a disorder). The seven-year-old boy may struggle because he jumps to answers before he has gathered enough information. His brain sets the bar for “enough evidence” too low. The thirty-two-year-old woman may find it hard to stay motivated because future rewards feel too far away. A project due next week carries little weight compared to what is happening right now, so effort gets deferred. The twelve-year-old boy may have normal decision-making machinery, but anxiety floods his system in stressful contexts, disrupting his ability to focus.

These are different problems, each with its own explanation and potentially its own solution. The central idea in computational psychiatry is that what we call ADHD is not one thing, but the outward expression of several underlying glitches in how the brain processes information, weighs rewards, and regulates effort.

Decision-making and information use. One line of research suggests that people with ADHD often set their internal decision threshold too low, meaning they act before they have enough evidence. They may also gather information less efficiently, making their decisions more variable and error-prone (Huang-Pollock et al., 2017).

These patterns can be measured. In a typical perceptual decision task, people watch dots moving on a screen and judge their overall direction. Some respond quickly but make errors; others wait longer and get it right. By analyzing the full pattern of response times and accuracy across many trials, drift-diffusion models estimate how efficiently someone integrates evidence (drift rate), how much evidence they require before committing (decision threshold), and how long it takes to initiate the response (non-decision time). Recent applications in child samples find both inattention and hyperactivity-impulsivity linked to reduced drift rates, with inattention also showing longer non-decision times, suggesting greater difficulty getting started on tasks (Ging-Jehli et al., 2024). Medication can shift decision parameters in DDM-modeled tasks, consistent with higher thresholds and faster evidence use (Pedersen et al., 2017).

From Ging-Jehli and Pine, 2025, Neuropsychopharmacology. You can read the article here.

Rewards and time. Another body of work focuses on how people value future outcomes. In a delay discounting task, someone chooses between smaller rewards available immediately and larger rewards available later ($10 today versus $20 in a month). People with steep discounting consistently choose the immediate option even when waiting would yield substantially more. This helps explain why sustained effort on long-term tasks feels so aversive for many with ADHD. The classic “dual-pathway” theory proposes that ADHD can arise either from these reward-processing issues or from executive function problems like working memory and inhibition (Sonuga-Barke, 2003). Later refinements added timing itself as a third pathway, since many with ADHD also struggle with judging and predicting intervals of time (Sonuga-Barke et al., 2010).

These computational problems may map onto specific brain circuits. Dorsal frontostriatal pathways linking prefrontal cortex and dorsal striatum support cognitive control and decision-making. Orbitofronto-striatal circuits connecting ventromedial prefrontal cortex and ventral striatum handle reward processing and temporal discounting. Fronto-cerebellar pathways mediate timing and temporal prediction (Durston et al., 2011). Dopamine and noradrenaline modulate these circuits, tuning parameters like decision thresholds, reward sensitivity, and the time course of learning (Arnsten & Pliszka, 2011). Stimulant medications may work by increasing catecholamine availability, which can normalize reward learning rates and reduce temporal discounting (Sethi et al., 2018; Shiels et al., 2008).

Seen this way, the three vignettes make more sense. The seven-year-old may have had noisy decision-making that improved with brain development. The adult woman may have a steep discounting of future rewards. The anxious twelve-year-old may have intact decision-making but disrupted thresholds in stressful contexts. These are testable ideas, and they point toward why one-size-fits-all categories struggle to capture the reality of ADHD.

Algorithmic Circuit Psychiatry

If ADHD aggregates multiple computational problems, each running on distinct neural circuits, this opens a path toward precision psychiatry through an algorithmic circuit framework. Instead of asking “Does this person have ADHD?” we could ask: “Which algorithms are altered, in which circuits, and what interventions might target them?”

Consider three hypothetical subtypes based on the computational framework above. Each generates specific predictions about brain circuits, developmental course, and treatment response.

A decision-noise subtype. People in this group set their decision thresholds too low and gather evidence inefficiently. They commit to responses before accumulating enough information. This maps onto frontoparietal networks that integrate sensory evidence and dorsolateral prefrontal cortex that sets response thresholds. If this is primarily a maturation problem in frontal systems, symptoms might improve substantially as the prefrontal cortex develops through adolescence. Stimulant medications, which raise decision thresholds, should help. Non-invasive brain stimulation targeting right dorsolateral prefrontal cortex, which has shown some promise in preliminary ADHD studies, might modulate threshold-setting directly (Alyagon et al., 2020; Rubia et al., 2021).

A temporal discounting subtype. People here struggle because future outcomes carry too little motivational weight. They can focus when rewards are immediate, but sustained effort toward distant goals feels aversive. This maps onto orbitofronto-striatal circuits, particularly connections between ventromedial prefrontal cortex and ventral striatum that represent the value of delayed rewards. Stimulant medications reduce temporal discounting (Shiels et al., 2008), and atomoxetine (a selective noradrenaline reuptake inhibitor) may work through similar mechanisms. Behavioral interventions that break long-term goals into shorter intervals with more frequent reinforcement should help. Neurostimulation targeting ventromedial prefrontal regions could theoretically modulate reward valuation, though this remains speculative.

A timing subtype. These individuals have difficulty with temporal processing itself, struggling to judge intervals, predict when events will occur, and coordinate actions in time. This maps onto fronto-cerebellar circuits. Timing deficits may persist even when attention improves, as timing and cognitive control appear to be dissociable (Sonuga-Barke et al., 2010). Methylphenidate improves timing performance in some studies, but the effects are less consistent than for attention or response inhibition. Cerebellar-targeted interventions, though largely unexplored in ADHD, represent a logical direction for this subtype.

These are hypotheses to guide future research. The subtypes may overlap, with individuals showing deficits in multiple domains. The circuit mappings are approximations, as brain systems are highly interconnected. And the intervention predictions remain largely untested. But this is precisely the point: the algorithmic circuit framework generates testable predictions that symptom-based diagnosis does not.

The Validation Challenge

This reframing is appealing, but does it actually solve the validity problem? Not yet. We still need to show that these computational subtypes are real and stable, not just convenient theoretical constructs. That means testing them the same way medicine tested whether “dropsy” was actually three different diseases.

If steep temporal discounting, noisy decision-making, and timing deficits are truly distinct subtypes, they should diverge on the validators. They should have different developmental courses. They should run in families in distinct patterns. They should predict different treatment responses. And they should map onto different patterns of brain structure and function.

The honest answer is that this work has not been done yet. The specific computational subtypes I outlined are hypotheses, not established entities. We have evidence that ADHD shows heterogeneity, and some studies suggest it can be subdivided in various ways. But no one has systematically tested whether decision threshold, temporal discounting, and timing represent distinct, validated subtypes using the full set of validators.

There is also a fundamental measurement problem. The tasks we use to assess computational parameters show poor reliability. Studies report intraclass correlations below 0.5, sometimes below 0.3, for measures of inhibition, working memory, and temporal discounting (Hedge et al., 2018). If you cannot measure something consistently, you cannot use it to predict outcomes or define subtypes.

But this problem is likely fixable. Schurr and colleagues (2024) showed in a longitudinal study that apparent unreliability in computational parameters may partly reflect real within-person variability driven by practice effects and emotional states. Hierarchical Bayesian models that account for trial-level variability and state-dependent changes can substantially improve parameter estimates (Katahira, 2016). With enough trials, measured over multiple sessions and analyzed properly, you can obtain stable individual-level estimates of computational parameters. These “behavioral fingerprints” could then be tested rigorously against the classical validators.

Moving Forward

What remains is to deploy these methods in prospective studies that track individuals from childhood through adolescence, measure treatment response, and test whether computational subtypes outperform symptom-based diagnosis in predicting outcomes. The computational tools exist: hierarchical Bayesian parameter estimation for stable individual estimates, large-scale task batteries to measure multiple algorithmic dimensions, and normative modeling to identify circuit substrates. The question is whether these tools, when properly applied, can decompose ADHD into validated entities the way bacteriology decomposed the fevers.

The validity debate has exposed the limits of purely descriptive diagnosis. For now, we do not need to conclude that ADHD is invalid or that treatment is futile. We need to recognize that the current category is provisional, a placeholder for phenomena we can now start to decompose into testable mechanisms. That requires updating our measurement tools and our validation frameworks, but let’s not discard the gains already made.

References

Alyagon U, Shahar H, Hadar A, Barnea-Ygael N, Lazarovits A, Shalev H, Zangen A (2020). Alleviation of ADHD symptoms by non-invasive right prefrontal stimulation is correlated with EEG activity. NeuroImage: Clinical 26:102206.

Arnsten AFT, Pliszka SR (2011). Catecholamine influences on prefrontal cortical function: relevance to treatment of attention deficit/hyperactivity disorder and related disorders. Pharmacology Biochemistry and Behavior 99(2):211-216.

Bush G, Spencer TJ, Holmes J, Shin LM, Valera EM, Seidman LJ, Makris N, Surman C, Aleardi M, Mick E, 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.

Caye A, Rocha TB, Anselmi L, Murray J, Menezes AM, Barros FC, Gonçalves H, Wehrmeister F, Jensen CM, Steinhausen HC, Swanson JM, Kieling C, Rohde LA (2016). Attention-Deficit/Hyperactivity Disorder Trajectories From Childhood to Young Adulthood: Evidence From a Birth Cohort Supporting a Late-Onset Syndrome. JAMA Psychiatry 73(7):705-712.

Durston S, van Belle J, de Zeeuw P (2011). Differentiating frontostriatal and fronto-cerebellar circuits in attention-deficit/hyperactivity disorder. Biological Psychiatry 69(12):1178-1184.

Ging-Jehli NR, Ratcliff R, Arnold LE, Huang-Pollock CL (2024). Applying mechanistic models to clinical research in ADHD: challenges and opportunities. Molecular Psychiatry 29:3542-3554.

Hedge C, Powell G, Sumner P (2018). The reliability paradox: Why robust cognitive tasks do not produce reliable individual differences. Behavior Research Methods 50:1166-1186.

Huang-Pollock C, Ratcliff R, McKoon G, Shapiro Z, Weigard A, Galloway-Long H (2017). Using the diffusion model to explain cognitive deficits in Attention Deficit Hyperactivity Disorder. Journal of Abnormal Child Psychology 45(1):57-68.

Karalunas SL, Gustafsson HC, Dieckmann NF, Tipsord J, Mitchell SH, Nigg JT (2017). Heterogeneity in development of aspects of working memory predicts longitudinal ADHD symptom change. Journal of Abnormal Psychology 126(6):774-792.

Katahira K (2016). How hierarchical models improve point estimates of model parameters at the individual level. Journal of Mathematical Psychology 73:37-58.

Kessler RC, Adler L, Barkley R, Biederman J, Conners CK, Demler O, Faraone SV, Greenhill LL, Howes MJ, Secnik K, Spencer T, Ustun TB, Walters EE, Zaslavsky AM (2006). The prevalence and correlates of adult ADHD in the United States: results from the National Comorbidity Survey Replication. American Journal of Psychiatry 163(4):716-723.

Luo L, He N, Geng H, et al. (2024). Mapping ADHD heterogeneity and biotypes through topological deviations in morphometric similarity networks. medRxiv preprint.

Molina BSG, Hinshaw SP, Swanson JM, Arnold LE, Vitiello B, Jensen PS, Epstein JN, Hoza B, Hechtman L, Abikoff HB, Elliott GR, Greenhill LL, Newcorn JH, Wells KC, Wigal T, Gibbons RD, Hur K, Houck PR; MTA Cooperative Group (2009). The MTA at 8 years: prospective follow-up of children treated for combined-type ADHD in a multisite study. Journal of the American Academy of Child and Adolescent Psychiatry 48(5):484-500.

Pedersen ML, Frank MJ, Biele G (2017). The drift diffusion model as the choice rule in reinforcement learning. Psychonomic Bulletin & Review 24:1234-1251.

Robins E, Guze SB (1970). Establishment of diagnostic validity in psychiatric illness: its application to schizophrenia. American Journal of Psychiatry 126(7):983-987.

Rouder JN, Haaf JM (2019). A psychometrics of individual differences in experimental tasks. Psychonomic Bulletin & Review 26:452-467.

Rubia K, Alegria AA, Cubillo AI, Smith AB, Brammer MJ, Radua J (2014). Effects of stimulants on brain function in attention-deficit/hyperactivity disorder: a systematic review and meta-analysis. Biological Psychiatry 76(8):616-628.

Rubia K, Westwood S, Aggensteiner PM, Brandeis D (2021). Noninvasive brain stimulation in children and adults with attention-deficit/hyperactivity disorder: a systematic review and meta-analysis. Journal of Psychiatry & Neuroscience 46(1):E14-E33.

Schurr R, Reznik D, Hillman H, Bhui R, Gershman SJ (2024). Dynamic computational phenotyping of human cognition. Nature Human Behaviour 8:917-931.

Sethi A, Voon V, Critchley HD, Cercignani M, Harrison NA (2018). A neurocomputational account of reward and novelty processing and effects of psychostimulants in attention deficit hyperactivity disorder. Brain 141(5):1545-1557.

Shaw P, Eckstrand K, Sharp W, Blumenthal J, Lerch JP, Greenstein D, Clasen L, Evans A, Giedd J, Rapoport JL (2007). Attention-deficit/hyperactivity disorder is characterized by a delay in cortical maturation. Proceedings of the National Academy of Sciences USA 104(49):19649-19654.

Shiels K, Hawk LW Jr, Lysczek CL, Tannock R, Pelham WE Jr, Spencer SV, Gangloff BP, Waschbusch DA (2008). The effects of methylphenidate on discounting of delayed rewards in ADHD. Experimental and Clinical Psychopharmology 16(5):411-419.

Sonuga-Barke EJ (2003). The dual pathway model of AD/HD: an elaboration of neuro-developmental characteristics. Neuroscience & Biobehavioral Reviews 27(7):593-604.

Sonuga-Barke EJ, Bitsakou P, Thompson M (2010). Beyond the dual pathway model: evidence for the dissociation of timing, inhibitory, and delay-related impairments in attention-deficit/hyperactivity disorder. Journal of the American Academy of Child and Adolescent Psychiatry 49(4):345-355.

Willcutt EG (2012). The prevalence of DSM-IV attention-deficit/hyperactivity disorder: a meta-analytic review. Neurotherapeutics 9(3):490-499.