Effects of a computerized working memory training program on working memory, attention, and academics in adolescents with severe LD and comorbid ADHD: a randomized controlled trial.

The study compared two computerized training programs head-to-head:

**Intervention group:** Cogmed RM (RoboMemo), a commercially available working memory training program. It consists of 25 sessions (30–45 minutes each) of adaptive visual-spatial and auditory-verbal working memory tasks. The difficulty adjusts automatically to keep the user at their limit — if you get a trial correct, the next one gets harder; if you get it wrong, it gets easier. This "adaptive" feature is the core of the intervention.

**Active control group:** Academy of MATH, a computerized math skills training program. This was not a placebo — it was a real educational intervention targeting arithmetic, math reasoning, and number concepts. Participants completed the same number of sessions (25) for the same duration.

**Outcome measures** were grouped into four categories:

- **Criterion measures** (tasks very similar to what was trained): Digit Span Forward/Backward (auditory-verbal short-term and working memory) and Spatial Span (visual-spatial working memory).

- **Near transfer measures** (different working memory tasks): CANTAB Spatial Working Memory (a more complex spatial memory task) and a listening span task.

- **Far transfer measures** (real-world outcomes): Teacher and parent ratings of inattention/hyperactivity (IOWA Conners Rating Scale), and academic achievement tests (Woodcock-Johnson III: reading comprehension, math calculation, spelling, writing fluency).

- **Compliance measures:** Number of sessions completed and the Cogmed Improvement Index (a built-in score tracking improvement on the training tasks themselves).

**Sample size:** 60 adolescents (52 male, 8 female)

**Age range:** 12–17 years (mean 14.3 years)

**Population:** Adolescents with severe, co-existing Learning Disabilities (LD) and Attention Deficit Hyperactivity Disorder (ADHD), attending a specialized residential school in Toronto, Canada

**Severity:** All had IQ > 80 (mean Working Memory Index = 85, which is below average). Academic scores were more than 2 standard deviations below the mean on standardized tests. 82% scored below the 25th percentile in all academic subjects. 72% scored below the 25th percentile on working memory.

**Medication:** 59 of 60 participants were taking long-acting stimulant medication for ADHD throughout the study (continued as usual)

**Setting:** Semi-residential school funded by the Ontario Ministry of Education, designed for adolescents with severely impairing LD and ADHD who had failed to respond to standard interventions in their home communities

**Auditory-verbal working memory:** Digit Span Forward (DSF) and Digit Span Backward (DSB) from the WISC-IV. Forward span measures short-term memory (holding digits in mind); backward span measures working memory (holding AND manipulating digits). Scores range 0–9 (highest span length correctly reproduced).

**Visual-spatial working memory:** Spatial Span (SSP) from the CANTAB battery. A computer screen shows 9 white squares that light up in sequence; the participant must reproduce the sequence (forward = short-term memory, backward = working memory). Scores 0–9.

**Complex working memory:** CANTAB Spatial Working Memory (SWM) — a self-ordered search task where participants find hidden tokens without revisiting locations they've already checked. Measures strategy and errors.

**Attention/behavior:** IOWA Conners Rating Scale — teacher and parent versions. Two subscales: Inattention/Overactivity (IO, 0–15 scale, higher = worse) and Oppositional/Defiant (OD, 0–15 scale). Also the Strengths and Difficulties Questionnaire (SDQ) Hyperactivity scale.

**Academic achievement:** Woodcock-Johnson III Tests of Achievement — reading comprehension, passage comprehension, word identification, word attack, spelling, calculation, math fluency, writing fluency. All reported as standard scores (mean = 100, SD = 15).

**Compliance:** Number of sessions completed (out of 25) and the Cogmed Improvement Index (automatically calculated by the software: Max Index minus Start Index).

**Study design:** Randomized controlled trial (RCT) with two parallel groups. Participants were randomly assigned to either working memory training (Cogmed RM) or math training (Academy of MATH).

**Randomization:** Participants were randomized using a computer-generated random number sequence. The flow of participants is shown in Figure 1 (see original paper): 61 adolescents were eligible, 60 consented and were randomized (31 to WM training, 29 to math training). One participant dropped out from the math group before starting. Final analysis included 30 in WM group and 29 in math group.

**Blinding:** The psychometricians (testers) who administered the outcome assessments were blind to group assignment — they did not know which training program each participant had completed. However, participants and teachers were not blind — they knew whether they were doing working memory games or math games. This is a significant limitation because expectation effects could influence behavior ratings and effort on tests.

**Duration:** The training lasted approximately 5–8 weeks (25 sessions, typically 5 sessions per week, 30–45 minutes per session). Assessments were conducted at baseline (pre-training) and within 3 weeks after completion (post-training). There was no long-term follow-up (e.g., 3 or 6 months later).

**Statistical approach:** The primary analysis used ANCOVA (Analysis of Covariance) comparing post-training scores between groups, controlling for baseline scores. This is appropriate for RCTs because it accounts for any pre-existing differences. Effect sizes were reported as partial eta-squared (η²p), which can be interpreted as: 0.01 = small, 0.06 = medium, 0.14 = large. They also examined correlations between training improvement (Cogmed Improvement Index) and changes in outcome measures.

**What this design can and cannot prove:**

**Can prove:** Because of random assignment, any differences between groups at post-test are likely due to the training programs themselves (not pre-existing differences). The active control group (math training) controls for non-specific effects like computer use, attention from staff, and time spent on task.

**Cannot prove:** The lack of a no-treatment control group means we cannot say whether either intervention is better than doing nothing. The lack of blinding for participants and teachers means that placebo effects or expectancy effects could influence results, especially on subjective measures like behavior ratings. The short follow-up (3 weeks) means we cannot assess whether any gains persist. The small sample size (N=60) means the study may have been underpowered to detect small-to-medium effects.

**Major methodological weaknesses:**

1. No sham/placebo control — math training is an active intervention, not a placebo

2. No blinding of participants or teachers (only testers were blind)

3. Short follow-up (3 weeks post-training, no long-term data)

4. Small sample size for detecting transfer effects

5. All participants were on stimulant medication, which could interact with training effects

6. The sample is highly specific (severe LD/ADHD in a residential school) — results may not generalize to less impaired populations

**Primary outcomes (criterion measures — working memory tasks similar to training):**

**Digit Span Backward (auditory-verbal working memory):** Significant group difference favoring WM training. WM group improved from mean 6.7 to 7.5; math group improved from 6.8 to 7.0. Effect size: η²p = 0.08 (medium). p = 0.03.

**Spatial Span Backward (visual-spatial working memory):** Significant group difference favoring WM training. WM group improved from mean 5.6 to 6.5; math group improved from 5.7 to 5.9. Effect size: η²p = 0.07 (medium). p = 0.04.

**Digit Span Forward (short-term memory):** No significant group difference (η²p = 0.01, p = 0.47).

**Spatial Span Forward (short-term memory):** No significant group difference (η²p = 0.00, p = 0.97).

**Secondary outcomes (near transfer — different working memory tasks):**

**CANTAB Spatial Working Memory (strategy score):** No significant group difference (η²p = 0.01, p = 0.50).

**CANTAB Spatial Working Memory (between errors):** No significant group difference (η²p = 0.01, p = 0.47).

**Listening Span (complex working memory):** No significant group difference (η²p = 0.00, p = 0.85).

**Secondary outcomes (far transfer — attention and behavior):**

**IOWA Conners Inattention/Overactivity — Teacher ratings:** No significant group difference (η²p = 0.00, p = 0.87).

**IOWA Conners Inattention/Overactivity — Parent ratings:** No significant group difference (η²p = 0.00, p = 0.95).

**IOWA Conners Oppositional/Defiant — Teacher ratings:** No significant group difference (η²p = 0.00, p = 0.99).

**IOWA Conners Oppositional/Defiant — Parent ratings:** No significant group difference (η²p = 0.00, p = 0.89).

**Secondary outcomes (far transfer — academic achievement):**

**Woodcock-Johnson Reading Comprehension:** No significant group difference (η²p = 0.00, p = 0.88).

**Woodcock-Johnson Math Calculation:** No significant group difference (η²p = 0.00, p = 0.96).

**Woodcock-Johnson Spelling:** No significant group difference (η²p = 0.01, p = 0.52).

**Woodcock-Johnson Writing Fluency:** No significant group difference (η²p = 0.00, p = 0.74).

**Exploratory analysis (within WM training group only):**

Among participants in the WM training group, those who showed greater improvement on the training tasks themselves (higher Cogmed Improvement Index) were rated by parents as less inattentive/hyperactive at post-test (r = -0.42, p = 0.03). This correlation was not found for teacher ratings.

**Digit Span Backward:** The WM training group improved by about 0.8 points (on a 0–9 scale), while the math group improved by about 0.2 points. This is a medium-sized effect (η²p = 0.08). In practical terms, this means the average participant in the WM group could hold and manipulate about one additional digit in their mind compared to the math group.

**Spatial Span Backward:** The WM group improved by about 0.9 points, the math group by about 0.2 points. Again a medium effect (η²p = 0.07). This translates to being able to reproduce about one additional spatial location in reverse order.

**No transfer effects:** For every other measure — attention ratings, academic tests, and more complex working memory tasks — the effect sizes were essentially zero (η²p = 0.00 to 0.01). This means the training had no detectable impact on real-world functioning.

**The correlation finding:** Among those who improved most on the training tasks (top performers), parent-rated inattention dropped by about 2–3 points on the IOWA Conners scale (0–15). This is a moderate correlation (r = -0.42), but it's correlational — we cannot say training caused the improvement.

**What the authors acknowledge:**

Small sample size (N=60) limits statistical power to detect small effects

No long-term follow-up (only 3 weeks post-training)

The active control condition (math training) may have produced its own benefits, making it harder to detect WM training effects

The sample is highly specific (severe LD/ADHD in a residential school) — results may not generalize

All but one participant were on stimulant medication, which could interact with training

**What a critical reader would note:**

**No blinding of participants or teachers:** This is a major issue. If teachers and parents knew which students were doing "brain training," they might have expected improvements and rated behavior more favorably. The fact that no effects were found on teacher ratings actually argues against this bias, but it remains a concern.

**No sham control:** The math training group received a real educational intervention. This is better than a no-treatment control, but it means we cannot rule out that both groups improved due to non-specific factors (attention, computer use, structured time).

**Multiple comparisons:** The study tested many outcomes (at least 15 primary and secondary measures) without correcting for multiple comparisons. The two significant findings (Digit Span Backward and Spatial Span Backward) could be chance findings. The authors do not report whether these survive correction.

**Short training duration:** 25 sessions over 5–8 weeks may be insufficient for transfer effects to emerge. Some studies suggest 50+ sessions may be needed.

**No measure of real-world working memory:** The criterion measures (Digit Span, Spatial Span) are very similar to the training tasks. Improvement on these may reflect practice effects rather than genuine cognitive enhancement.

**Industry funding:** Cogmed America provided the software licenses free of charge. One author (Tannock) had consulting relationships with pharmaceutical companies. While conflicts were disclosed, this is worth noting.

**Attrition:** One participant dropped out from the math group. The authors do not report intention-to-treat analysis (analyzing everyone as randomized, regardless of completion).

For someone running their own n=1 experiment:

**What to test:**

Test whether adaptive computerized working memory training (like Cogmed, or free alternatives like Dual N-Back) improves your working memory, attention, or cognitive performance.

The specific dose tested here: 25 sessions, 30–45 minutes each, 5 days per week, for 5–8 weeks. Each session includes 8–12 different working memory tasks that adapt to your performance level.

**Minimum meaningful duration:**

At least 25 sessions over 5–8 weeks. The study found improvements on trained-like tasks at this dose, but no transfer. If you want to test for transfer effects, you may need 50+ sessions (10+ weeks).

A shorter duration (e.g., 2 weeks) is unlikely to produce meaningful changes.

**What to measure:**

**Primary outcome:** Performance on working memory tests that are DIFFERENT from the training tasks. Use free online tests like:

- Digit Span (forward and backward) — available on many cognitive testing websites

- Spatial Span (Corsi block test) — free apps available

- Operation Span or Reading Span (complex working memory) — available on cognitive testing platforms

**Secondary outcomes (transfer):**

- Attention: Use a sustained attention test (e.g., SART or CPT) — free online versions exist

- Academic/cognitive: Reading comprehension (timed passage reading with questions), math calculation speed (e.g., 100 simple arithmetic problems timed)

- Self-report: Daily ratings of focus, distractibility, mind-wandering (1–10 scale)

**Compliance:** Track number of sessions completed and your improvement on the training tasks themselves (most programs provide a progress score)

**Key confounds to control for:**

**Practice effects:** Working memory tests improve with repeated testing. Use alternate versions of tests at pre and post, or test multiple times before starting training to establish a stable baseline.