Can Executive-Function Complaints and Quality of Life Predict Adult ADHD Stimulant Response? A 36-Patient Study

Short answer: A small naturalistic follow-up study found that baseline self-reported executive-function difficulty and higher quality-of-life ratings were associated with being classified as improved after stimulant treatment. In 36 medication-naïve adults with ADHD, a selected two-questionnaire logistic model reported odds ratios of 1.13 for the BRIEF-A Global Executive Composite and 1.23 for the Quality of Life Enjoyment and Satisfaction Questionnaire, with a within-sample area under the receiver-operating-characteristic curve of 0.81. These exploratory associations were not externally validated as an individual prediction rule. Neither a prescribed dose nor a questionnaire score established a chemically assayed amphetamine or methylphenidate concentration in legally collected human venous whole blood. The study also did not test intoxicated drivers or causal responsibility for an incident.

The original empirical source is DiSalvo, Biederman, O’Connor and colleagues (2026), Clinical predictors of stimulant efficacy in adults with ADHD, Therapeutic Advances in Psychopharmacology 16, doi:10.1177/20451253261428807. A central distinction is that the authors examined clinical response under ordinary management, not blood toxicology or a controlled head-to-head drug trial. The results are most informative when their positive exploratory model, negative baseline comparisons, treatment changes, and important overlap with an earlier brain-imaging report are kept visible together.

What scientific question did the investigators ask?

Rather than starting with an analytical method or a measured stimulant concentration, the investigators asked whether pretreatment ratings of everyday symptoms, psychiatric difficulties, executive functioning and quality of life were associated with subsequent clinician-rated improvement. ADHD treatment selection can require adjustments in formulation and dose. The practical attraction of inexpensive questionnaires is that they might eventually identify patients who need closer follow-up or different treatment strategies. But identifying an association inside a small discovery sample is only the first step toward establishing a usable predictor.

The study’s named outcome was the Clinical Global Impression–Improvement scale (CGI-I), not a laboratory attention task or a direct pharmacodynamic measure. A clinician’s judgment that a patient improved during several months of treatment can reflect medication, clinical follow-up, accommodations, concurrent health changes, expectancy, and other circumstances. The study could describe which baseline measurements tracked that outcome; it could not cleanly separate those mechanisms or estimate an individual’s chemical drug exposure.

How many independent adults were included?

Forty medication-naïve adults referred to the Massachusetts General Hospital adult ADHD program were initially included. Four lacked the required follow-up information and were excluded, leaving 36 independently analyzed people. The final sample was split equally: 18 individuals whose endpoint CGI-I rating met the responder criterion and 18 whose rating did not. Individuals attended an initial evaluation and two follow-up clinical visits; those repeated observations do not create 72 or 108 independent participants.

The endpoint-response split was determined after treatment was observed. It was not a random assignment to biological categories. Each patient contributed their own baseline questionnaire data, clinical history and later improvement classification. The study was relatively small and came from one referral program, so imprecision, referral selection, correlated questionnaires and a few influential observations can matter substantially. Differences between two groups of 18 should not be recast as universal response probabilities for adults with ADHD.

What does “responder” mean here?

The study classified a responder as someone with an endpoint CGI-I score of 1 or 2, corresponding to clinician-rated substantial improvement. Endpoint scores above 2 defined nonresponse for this analysis. This dichotomy is a convenient analytical choice; it does not mean that a person scoring 3 had no improvement, no symptomatic benefit or a pharmacologically inactive drug. CGI-I measures perceived change compared with pretreatment functioning; it is not an assay for absorbed stimulant or neurotransmitter release.

This distinction also matters when comparing other publications involving similar participants. A separate diffusion-imaging pilot from the same research group examined MRI fractional anisotropy and percentage improvement in the CGI-Severity scale. The cohort appears to overlap substantially with the present 36-patient sample, but the principal exposure measures and response definitions differ. Those papers should not be counted as independent replications of the same biomarker or presented as two independently recruited groups of 36.

Which stimulant drugs and doses were used?

Clinicians prescribed methylphenidate-family or amphetamine-family treatment at therapeutic doses as part of ordinary management. The original study explicitly states that actual dose information was not analyzed. Neither a uniform salt formulation nor a single fixed route-and-dose schedule was experimentally imposed and maintained across all participants. Changing preparations or switching drug families was part of the observed care pathway. It would therefore be inaccurate to invent a universal daily milligram amount or to call the exposure specifically methamphetamine, dextroamphetamine, lisdexamfetamine, or a particular extended-release product without a participant-specific source record.

At the baseline visit, 25 patients were recorded in the methylphenidate category and 11 in the amphetamine category. By the endpoint, the corresponding counts were 14 and 22. Those changing distributions are real treatment chronology, not randomly assigned parallel drug arms. A higher percentage of endpoint responders taking one category cannot establish comparative efficacy when switching was itself influenced by earlier outcomes.

How long was clinical follow-up?

The study followed patients for a mean of 116.1 days, with a substantial standard deviation of 77.9 days. Patients were evaluated at intake and twice afterward, rather than sampled at a series of identical postdose blood-draw intervals. This is a clinical management timeline, not a pharmacokinetic curve. It cannot be used to calculate peak amphetamine or methylphenidate levels, individual half-lives, the last dose clock-time, or the concentration that would have been present during a separate alleged event.

Variation in follow-up duration can influence response classification and opportunities for dose optimization. Clinician changes may have occurred because of efficacy, side effects or patient needs. Those features improve real-world relevance in one sense, but reduce the control needed for isolating a single medication’s causal effect.

Which baseline instruments were studied?

The baseline battery included the Adult ADHD Self-Report Scale, ADHD Rating Scale, Adult Self-Report (ASR), Behavior Rating Inventory of Executive Function–Adult Version (BRIEF-A), Social Responsiveness Scale-2, a Barkley emotional self-regulation scale, the Mind Wandering Questionnaire, and the Quality of Life Enjoyment and Satisfaction Questionnaire (Q-LES-Q). Different questionnaires measure different constructs, with different scoring directions. A high BRIEF-A score indicates more reported executive-function problems; a high Q-LES-Q score indicates better reported quality of life.

None is a chemical assay. A BRIEF-A Global Executive Composite does not count dopamine molecules, a Q-LES-Q total does not measure parent drug in a specimen, and an ASR thought-problems score does not by itself establish clinical psychosis or a particular person’s credibility. Keeping instrument names and constructs distinct prevents misleading conversions between clinical descriptions and forensic proof.

What did unadjusted group comparisons show?

Before the multivariable model, responders and nonresponders differed nominally on several baseline measures. Responders reported greater problems on BRIEF-A Working Memory and Task Monitoring, but had fewer ASR-reported intrusive, thought, withdrawn and internalizing problems. They also reported more inattentive and total ADHD symptoms on the ADHD Rating Scale. Some conventional p values were below .05, including Task Monitoring at .004 and Withdrawn at .01.

Crucially, the investigators separately corrected bivariate comparisons for multiple tests within questionnaire domains. They state that none of the initially nominally significant bivariate outcomes survived the specified Bonferroni thresholds. The new article must preserve that finding rather than count every exploratory p value below .05 as a confirmed independent predictor. Nominal significance, correction for multiple testing, and model selection answer different inferential questions.

Evidence table 1: actual baseline comparisons and genuine nulls

Values come from the original primary Table 1. This reorganization emphasizes genuinely tested nonsignificant comparisons rather than turning them into imaginary evidence of no biological relationship. The Q-LES-Q and overall BRIEF-A composite are particularly instructive: neither alone met the conventional .05 bivariate threshold in this small sample, yet both were retained in the later selected logistic model.

What did the final two-variable model report?

After entering variables with bivariate p values no greater than .10, the investigators used backward stepwise logistic regression to retain a BRIEF-A Global Executive Composite (GEC) and Q-LES-Q total-score model. The odds ratio for BRIEF-GEC was 1.13 per source-defined scale point, with a 95% confidence interval 1.02–1.25 and p = .019. The Q-LES-Q odds ratio was 1.23, interval 1.05–1.45, p = .010. The source reported a joint model p value of .003.

The signs of these associations are not contradictory: more executive problems and better quality of life could both coexist in a person, and the model estimates each association conditional on the other. Yet selection of variables in only 36 patients creates a real risk of overfitting. Odds ratios from a selected exploratory model should not be treated as stable population effects or applied directly as legal probabilities of treatment success.

Evidence table 2: final model, in-sample fit and treatment switching

The original authors described Nagelkerke pseudo-R² as 36.3% and reported an AUC of .81. Neither statistic establishes out-of-sample predictive accuracy or a biochemical mechanism. The model was built and assessed in the same small cohort; an independent cohort and a prespecified model would be needed before clinical deployment.

Does an AUC of 0.81 validate individual prediction?

No. AUC describes how well a score ranks observed responders above observed nonresponders in a defined dataset. When predictors are selected and performance is evaluated using the same 36 people, apparent discrimination can be optimistic. The paper did not report independently recruited external validation, a prespecified probability-calibration study, individual false-positive and false-negative rates suitable for clinical use, or a test that proves the formula transfers to another healthcare setting.

A model can distinguish two outcome groups in development data while performing much less effectively when applied prospectively. Small sample size, correlated baseline scales, multiple comparisons, treatment changes and variable follow-up make that concern substantive. An AUC is not the probability that a named patient will respond to a particular drug, and it is certainly not a probability that a person was impaired at an unrelated time.

Why did quality of life matter in the model?

The Q-LES-Q asks about enjoyment and satisfaction with everyday life. Its higher total score indicates a more favorable reported quality of life. The source found that, after accounting for the selected executive-function composite, higher baseline Q-LES-Q scores were associated with greater odds of the chosen clinician-rated response category. This could reflect behavioral supports, symptom burden, engagement with care, or unmeasured differences rather than a direct action of methylphenidate or amphetamine on a biochemical pathway.

The source does not establish that raising an individual’s Q-LES-Q score would pharmacologically increase their treatment responsiveness by 23%. Intervention causality cannot be inferred from a one-point observational odds ratio. The bivariate group difference itself was p = .06, a true nonsignificant result, highlighting the distinction between an unadjusted comparison and a conditional selected model.

Why could worse executive complaints accompany better response?

BRIEF-A GEC summarizes patient-reported executive-function difficulties, with higher scores indicating greater impairment. Participants with greater baseline difficulty may have had more room for observable change, or clinicians may have perceived larger improvement when marked everyday deficits lessened. Another possibility is measurement overlap, regression toward average ratings, coexisting support or adherence differences, and correlated clinical covariates. The research did not experimentally manipulate executive dysfunction to test these mechanisms.

Subscales also should not be conflated. A prior imaging article from this likely overlapping research cohort reported an association involving the BRIEF-A Working Memory subscale and imaging-derived fractional anisotropy, a different explanatory focus and underlying DOI. The present study instead identified a global executive composite and quality-of-life pair in an outcome-response logistic model. Neither represents directly measured cognitive-task performance, an independently replicated treatment algorithm, or quantitative parent-drug evidence.

Did other symptom scales independently establish prediction?

No final validated classifier was established for every individual measure. In the initial bivariate analysis, responders reported fewer ASR intrusive, withdrawn, thought and internalizing problems, whereas the ASR aggressive-behavior comparison was effectively unchanged (p = .95). Baseline Social Responsiveness Scale measures, mind wandering and emotional dysregulation also were not significant discriminators under the reported tests. It would be misleading to declare autism traits, depression, psychosis, impulsivity or misconduct from a single self-reported ASR subscale or a nonsignificant group comparison.

The investigators’ multiple-comparison correction is particularly important: several attractive unadjusted subgroup results were not robust under their own adjusted thresholds. These analyses may be useful for generating future hypotheses about clinical complexity, but the observed values do not establish independent causal determinants of medication response, and they cannot validate individual witness credibility or criminal behavioral predictions.

What did the sensitivity analysis add?

The authors examined whether baseline ADHD symptom ratings moderated the association between response status and BRIEF-GEC. An exploratory interaction was reported at p = .04 in one analysis, and the relation between ADHD rating severity and BRIEF-GEC was positive in both response groups. However, when the interaction was included in the variable-selection model, it failed the retention criterion and was removed. The final two-variable model remained the same.

That distinction should not disappear in secondary summaries. It is not correct to promote a nominal sensitivity-analysis interaction as a confirmed component of the final response algorithm. It is also not correct to assign a numerical p value to an unreported effect or treat a removed model term as proof of zero biological interaction. The exploratory analysis illustrates the fragility of small-sample modeling choices.

Do the endpoint drug-category results prove methylphenidate superiority?

No. Among the 14 people taking methylphenidate at the endpoint, 10 were responders; among the 22 taking an amphetamine-family medication, eight were responders. That unadjusted category comparison had p = .04. But the direction of treatment assignment and subsequent switching was not randomized or blinded: more nonresponders changed stimulant family during observation. In fact, 56% of nonresponders switched compared with 11% of responders, p = .005.

This is classic confounding by indication and treatment adaptation. Patients who had difficulty improving were more likely to receive a different drug family, meaning their endpoint exposure partly reflects their prior clinical trajectory. No causal conclusion about intrinsic chemical superiority follows from treating the endpoint category as if it had been randomly assigned at baseline. The study also did not analyze dose equivalence or steady-state exposure for the medication families.

What were the actual physical drug measurements?

There were no reported quantitative confirmatory drug measurements. The investigators collected questionnaire scores and clinician-rated outcomes; their clinical drug categories were prescribing information, not chromatography results. They did not report a concentration of methylphenidate, ritalinic acid, amphetamine, methamphetamine or a resolved d-/l-amphetamine enantiomer in legally collected human venous whole blood. They also did not quantify those drugs in separated plasma or serum, urine, oral fluid, sweat, hair, vitreous humor, milk, tissue, or breath.

Consequently there is no source analytical instrument, confirmatory analyte transition, specimen-specific calibration, limit of detection, limit of quantification, or concentration unit such as ng/mL for a measured parent drug. The absence of these data is an unmeasured endpoint, not a negative toxicology finding. One cannot infer that the patients were drug-free or had a specific concentration merely because the paper reported no quantitative assay.

Why distinguish venous whole blood from plasma and serum?

Whole blood contains cellular and liquid fractions, whereas plasma and serum are separated materials. Their drug concentration relationships depend on the compound, collection, preparation and analytical method. Even if a study had reported a valid plasma concentration, it could not automatically be relabeled a whole-blood result. Here, no corresponding stimulant concentration was reported in either compartment, so there is nothing valid to convert.

The separately published biological-matrices guidance explains why the specimen and purpose of the test must remain attached to its result. This clinical paper does not provide an individual toxicological concentration, a pharmacokinetic course, or an evidentiary method for reconstructing a legally relevant blood value from symptom reports.

Could the prescribed dose or response score reveal drug exposure?

No. A prescription category documents a clinician’s treatment choice, not ingestion at a verified time or absorption of a known fraction. Patients may differ in administration, adherence, formulation, dose adjustment and metabolism. A CGI-I response score adds an assessment of clinical change but no chemical identity or concentration. The study did not analyze the actual dose records, so substituting a customary dose or a presumed therapeutic concentration would introduce information absent from the original evidence.

Forensic confirmatory testing requires a valid specimen, known collection time, validated identification and quantification procedure, and appropriate quality controls. The separate confirmatory drug-testing explanation discusses that analytical workflow. A psychometric association cannot stand in for a chemical analysis, nor can it supply a legal numerical cutoff or reliable clock for last consumption.

What does the study say about impairment or driving?

It measured neither. Participants were adult ADHD patients undergoing treatment follow-up, not drivers performing a validated road test, persons involved in crashes, or witnesses assessed during an alleged incident. They were not evaluated for contemporaneous drug intoxication or driving impairment with paired venous whole-blood samples. The study therefore cannot establish that a given blood level caused or prevented erratic driving, that a particular patient was impaired on a specific date, or that a witness spoke accurately or inaccurately because of medication.

Group-level associations among questionnaires and clinician ratings can inform hypotheses about care. Case-specific medico-legal opinions require different evidence: authenticated chronological records, actual specimens where relevant, validated analytical results, observed function, co-exposures, medical context, and transparent limitations. Even with those sources, individual causation may remain uncertain.

How should the study’s statistical limitations be reported?

The small final sample of 36, removal of four initial recruits, recruitment from one clinical program, selection of predictors from multiple scales, backward elimination, treatment switching, variable follow-up duration and absent drug-dose analysis all restrict generalization. Confidence intervals around the two odds ratios give useful sampling uncertainty, but cannot correct unmeasured confounding or model-selection optimism. Likewise, the pseudo-R² and apparent AUC are descriptions of model fit in development data, not randomized causal effects or validated individual predictions.

Neither a nonsignificant bivariate p value nor a significant selected-model p value should be silently rewritten as the other. The source’s Bonferroni result is a real caution: none of the initially nominally significant questionnaire contrasts passed its domain-specific correction. The best framing is that investigators found a promising but preliminary multivariable association requiring independent, prospective validation.

What is the defensible forensic conclusion?

In one small naturalistic ADHD cohort, pretreatment self-reported global executive difficulty and better quality of life were retained in a selected model associated with later clinician-rated improvement. The model’s reported AUC of .81 was not externally validated, questionnaire screening was sensitive to multiple-comparison correction, and endpoint drug-family comparisons were confounded by clinical switching. The same dataset was not a blood-drug assay, controlled pharmacokinetic trial, or assessment of any person’s driving or incident causation.

The source-faithful conclusion is therefore narrower than a claim that questionnaire scores can determine which stimulant an individual will benefit from or prove chemical exposure. The original empirical analysis can be checked directly at DiSalvo M, Biederman J, O’Connor H, et al., Clinical predictors of stimulant efficacy in adults with ADHD, Therapeutic Advances in Psychopharmacology (2026), original study and DOI.

Frequently asked questions

1. How many independent people, responders and nonresponders were studied?

Thirty-six independently analyzed adult patients, including 18 with endpoint CGI-I scores of 1 or 2 and 18 with higher scores. Forty originally entered and four were excluded because complete follow-up data were unavailable. Baseline and follow-up visits are repeated observations within those people, not additional independent biological subjects.

2. Did the two-questionnaire model predict response accurately in new patients?

Not established. The source reported AUC .81 for a model built in the 36-person sample, but did not provide independent external validation, prospective calibration, or a transportable individual classification error estimate. Small-sample backward selection can produce optimistic performance.

3. Why does higher BRIEF-GEC associate with response despite indicating more problems?

BRIEF-GEC reflects greater self-reported executive-function difficulty when higher. The selected model’s positive association may reflect baseline severity, opportunity for improvement or unmeasured clinical factors; it does not show that causing more executive difficulty improves stimulant response. Q-LES-Q scores point in the opposite scoring direction: higher means better quality of life.

4. Were baseline executive and psychiatric group comparisons significant after correction?

No. Although several unadjusted comparisons were nominally below p=.05, the authors report that none passed their domain-specific Bonferroni corrections. Some initial comparisons, including BRIEF-GEC at p=.07 and Q-LES-Q at p=.06, were also nonsignificant before multivariable selection. The selected model is distinct and exploratory.

5. Did the study establish methylphenidate is more effective than amphetamine?

No. Endpoint responder frequencies differed, but more nonresponders had switched medication families, and dose information was not analyzed. The treatment pathway was naturalistic rather than a blinded, dose-matched randomized comparison. The source does not support causal superiority of either family or a universal formulation-specific effect.

6. Were amphetamine or methylphenidate concentrations measured in legal human venous whole blood?

No. The paper used prescription categories, questionnaires and clinician ratings, not a reported confirmatory chemical assay of parent stimulants or metabolites in human whole blood, plasma, serum or other matrices. It provides no specimen-based drug concentration, analytical limit, individualized last-use estimate or validated forensic threshold.

7. Can these clinical predictors determine whether an identified driver was impaired or caused a collision?

No. The study did not measure driving, collisions, intoxication or eyewitness performance, and did not pair tested function with blood drug levels. Clinical response associations cannot be converted into an individual legal conclusion about impairment, credibility, injury or incident causation without independently appropriate evidence.