IHAMS (NCT-01165463) was reviewed and approved by the University of Iowa (IRB Protocol 200908789), and funded by the NIH (RC1 AG035546). Written informed consent was obtained from all participants at baseline, and verbal consent was obtained at all follow-up interviews.

The protocol for this trial and supporting CONSORT checklist are available as supporting information (see Checklist S1 and Protocol S1). Five adjustments were made to the funded protocol prior to enrollment. First, direct randomization to four study arms within age bands was used, rather than initial randomization to three study arms with subsequent second level randomization to no booster vs. booster training within the on-site visual speed of processing training arm. Second, UFOV was specified as the sole primary outcome because no data existed from ACTIVE or any other study for the other neuropsychological outcomes, which were re-designated as secondary outcomes. Third, the power calculations were revised based on additional analyses of the one-year UFOV results from the ACTIVE study. Fourth, participants were trained in two-hour sessions, rather than the one-hour sessions used in ACTIVE due to the logistical and financial constraints of the funding mechanism. Finally, Blom rank transformations were used within age bands for normalization of the left and right censored neuropsychological outcomes, with linear mixed models used for consistency with prior studies.

Details about the protocol [32] and post-training results on the primary outcome (UFOV) [33] have previously been reported. Figure 1 presents the IHAMS CONSORT flow diagram for the three visual speed of processing active intervention arms (commercially available as Road Tour from Posit Science Corporation, San Francisco, CA, USA) vs. the attention control computerized crossword puzzle game (commercially available as Boatload of Crosswords, Boatload Puzzles, Yorktown Heights, NY, USA).

Sample size estimation was based on a secondary analysis of the ACTIVE data where the average one-year improvement in the UFOV score for the no contact controls was 69.9 ms (standard deviation = 159.1 ms) vs. 220.5 ms (standard deviation = 188.3) for the speed of processing group, for a 150.6 ms improvement differential. With 650 participants overall, a minimum of 138 in each group after 15% attrition by the one-year follow-up, and a reduced improvement differential of only 60 ms (due to IHAMS' inclusion of younger participants, use of an attention control group, and unconditional randomization to booster training), there would be 81.6% power (alpha = 0.05, two-tailed) for the primary outcome.

Potential participants were identified from patients attending the general internal or family medicine clinics at the University of Iowa. Using the electronic medical record (Epic, Verona, WI, USA), initial inclusion criteria were age (≥50), having≥2 visits to the clinic in the past year, and having no ICD9-CM codes indicating cognitive impairment. Cognitive impairment codes were Alzheimer's disease (331.0), Pick's disease (331.1), arteriosclerotic dementia (290.4 to 290.43), other senile or pre-senile dementia (290.0 to 290.9), dementia due to alcohol (291.1 to 291.2) or drugs (292.82 to 292.83), amnestic syndrome (294.0), or dementia due to other organic conditions (294.1). This electronic screening identified 5,743 potential participants.

Randomly selected (without replacement) weekly samples of 250 of these 5,743 individuals were then sent letters beginning in February 2010 and ending in November 2010 that were signed by the clinic medical directors and the principal investigator (FW) inviting them to participate by calling the IHAMS project office. Responding individuals underwent a brief screening telephone call to determine whether they met further inclusion criteria that could not be assessed using the electronic medical record. These were (a) having a PC and internet connection in the home, (b) no significant uncorrected self-reported vision issues, (c)<3 errors on the Short Portable Mental Status Screening Questionnaire [35], and (d) living within 30 miles of the project office. Of the 5,743 potential participants, 1,627 could not be reached, 2,079 did not respond or declined, 390 failed to meet one or more of the inclusion (telephone screening) criteria, and 966 had not been sent their letters by the time study enrollment closed.

The 681 eligible participants were invited to the project laboratory where written informed consent was obtained prior to their enrollment and baseline assessments, which were conducted by trained interviewers. Participants were then randomized by the study statistician (MJ) within age bands (50–64 and≥65) using a computerized algorithm, a 3∶3∶4∶4 allocation ratio, random permuted blocks of 4, 8, and 12, and sequentially numbered opaque envelopes [32], [33]. Because the participants' group assignments remained sealed until after the end of their initial visits, when the envelope was opened by the study coordinator (MD), the baseline assessments were fully double-blinded. After that, however, participants were no longer blinded because they knew the group to which they had been assigned.

After their baseline assessments and randomization, participants were invited back to the project laboratory on a subsequent day for their first training session. At that first session each on-site training participant received a scripted 15-minute introduction on how to use the assigned training program before going to one of two identical training rooms (one had the visual speed of processing program on five private work stations, while the other had the attention control program on five private work stations) to complete the first of their five weekly, two-hour training sessions. After completing 10 hours of training (the same basic training dose used in ACTIVE), or at approximately 6–8 weeks post-randomization for non-adherents (whichever came first), all on-site participants were invited back to the project laboratory for post-training assessments. One of the on-site active intervention training groups was also invited back at 11 months for two additional 2-hour “booster” training sessions (as in ACTIVE). All on-site participants were invited back for one-year assessments at which blinding was broken for the assessors but not for the investigative team at the concluding set of questions that solicited the participants' views about their assigned training program. After completing the one-year assessments, all on-site participants were given a copy of the visual speed of processing training software and installation directions to take home and load on their PCs and use at their discretion in perpetuity.

The at-home training participants were also invited back to the research lab after their baseline assessments, and received the scripted 15-minute introduction on how to use the visual speed of processing training program. This was followed by a 5–10 minute scripted introduction on how to load the training software onto their home PC. These participants were then given a copy of the training software and the loading instructions and asked to go home, load it on their PCs, and then complete 10 hours of training. Like the patients in the other three training groups, the at-home training participants were invited back for their post-training assessments at 6–8 weeks post-randomization, and for their one-year assessments. After completing the one-year assessments, at-home participants were told that they could continue to use the visual speed of processing training software on their home PCs at their discretion in perpetuity.

Figure 2 shows the appearance of the training program. After clicking on the start button in Figure 2a, Figure 2b is shown where the license plate and the eight circular locations surrounding it are masked. The masked license plate is then replaced (Figure 2c) with the target vehicle, either a car or a truck, and the eight masked circular locations are replaced with seven distracter stimuli (rabbit crossing signs in this example) or the target stimulus (always the Route 66 sign). The stimuli signs are presented for a specified time and are then replaced by Figure 2d. The amount of time that Figure 2c remains on the screen is measured in milliseconds (ms). In Figure 2e, both target vehicles (the car and truck) and the 8 circular locations are presented and the user is first required to select the correct target vehicle before selecting the circular location where the target sign appeared (Figure 2f).

Cognitive processing speed is trained by progressively reducing the ms of exposure that Figure 2c must remain on the screen for correct identification of both stimuli. The training program does not advance to a greater challenge until participants can maintain a 75% success rate at the current challenge level. As performance improves the challenge level is increased as follows: the visual field expands to add medium and distal orbits, these are accompanied by an increasing number of distracters (up to 47), and the vehicle pairs morph through nine different stages to become more similar and thus increasingly difficult to differentiate.

The attention control group used a computerized crossword puzzle game for training purposes. This game offered a choice between three puzzle sizes, three levels of complexity, and varying font sizes. It also included optional help features such as filling in an unknown letter or word. There is, however, no progressive challenge to the user by increased speed, visual field size, number of distractors, or degree of difficulty of target stimulus differentiation. Thus, the crossword puzzle program provides an appropriate, computerized attention control alternative to the active visual speed of processing training.

The primary outcome [33] was the UFOV [19], which was obtained at the baseline, post-training, and one-year assessments. It includes stimulus identification, divided attention, and selective attention subtests scored from 17-500 ms reflecting the shortest exposure time at which the participant could correctly perform each subtest, with a composite score ranging from 51–1500 ms. The secondary outcomes [33] were obtained at the baseline and one-year assessments. They were the Trail Making A and B tests (Trails [36]), the Symbol Digit Modalities Test (SDMT [37]), the Stroop Color and Word Test (Stroop [38]), the Controlled Oral Word Association Test (COWAT [39]), and the Digit Vigilance Test (DVT [40]). The Trails assess visual scanning ability, processing speed, and set-shifting/executive functioning, and are coded as the number of seconds to correctly complete connecting the number (Trails A) and number-letter (Trails B) sets. The SDMT is a substitution task that captures divided attention and processing speed, and is based on how many of 110 possible digit-symbol pairs are correctly scored in 90 seconds. The Stroop assesses the effect of interference with and inhibition on cognitive flexibility and processing speed, and other aspects of executive functioning, and is scored as the correct number of words, colors, and color-words identified in 45 seconds on each subtest. The COWAT is a phonemic test that assesses verbal fluency based on the number of unique words beginning with the letter C (or F or L) generated by the participant during 60 seconds, with a composite score of the number of correct words across the three letter trials. The DVT is a visual cancellation task that assesses sustained attention and psychomotor speed, and is performed by crossing out randomly placed 6′s in 59 rows of numbers, and is scored as the error (commission and omission) and time totals.

Means on the selected background characteristics and the neuropsychological outcomes were compared by intervention group and age band using chi-squared, analysis of variance, or Kruskal-Wallis tests for independent samples, as appropriate. The distributions on the neuropsychological tests are typically not normal due in part to left and right censoring (i.e., fixed minimal and maximal scores), and this was the case in IHAMS. At baseline, four of the outcomes were positively skewed (skewness = 1.3 to 3.0), six outcomes were leptokurtic (kurtosis = 1.3 to 7.1), and both distributional violations occurred for four outcomes (UFOV, Trails A, Trails B, and DVT errors). Therefore, Blom rank transformations were used to improve distributional normalization for model estimation, which resulted in means of zero and standard deviations of one [41], and yielded results that are directly interpretable as standardized effect sizes (Cohen's d [42]) because the effect of each intervention group vs. the attention control group is now measured in standard deviations. The Blom rank transformations were calculated separately within age bands to adjust for the different initial ability levels and to facilitate relative effect size comparisons.

Although attrition was modest (9%), it was evaluated using multivariable logistic regression (with backwards elimination given the small number of attriters) on the baseline neuropsychological tests, the treatment groups, age bands, and self-reported health [43]. Intent-to-treat analyses among the 620 (91%) respondents with complete data at the baseline and one-year assessments were conducted using linear mixed models. These models included between-person main effects for treatment group and age band, within-person main effects for time, and all two- and three-way interaction terms. Dunnett tests were used to compare each visual speed of processing training group to the attention control group. Significant observed effects were then converted into years of protection against age-related declines and/or years of improvement in cognitive performance by re-estimating the models using the original scalar values (i.e., non-Blom rank transformed scores), and dividing those effects by annualized age declines determined from prior normative studies. To address non-adherence to training (and provide a preliminary effectiveness analysis), especially within the at-home training group, the analyses were repeated after restricting the analytic sample to those who completed at least one hour of training.

91% of the participants completed their one-year assessments. To gauge the extent of differential attrition, the baseline neuropsychological tests, treatment groups, age bands, and self-reported health were all considered in an initial multiple logistic regression model. Only three of these variables were retained on the final step of the backwards elimination process. Participants in the booster training group had greater odds of being lost to follow-up (adjusted odds ratio [AOR] = 2.051, p = 0.021; the adjustment was for self-rated health and errors on the DVT test), as did those with lower self-reported health (AOR 1.597 per lower category rating, p = 0.004; the adjustment was for the booster training group status and errors on the DVT test), and those with more errors on the DVT test (AOR = 1.040 per error, p = 0.006; the adjustment was for booster training group status and self-rated health). Overall, the model fit the data reasonably well, with a p value of 0.710 on the Hosmer-Lemeshow statistic, and an area under the curve (AUC) of 0.690.

Median training time was 13 hours for the on-site visual speed of processing training with boosters group vs. 9–10 hours for the other training and attention control groups. These average training times, however, mask substantial heterogeneity in adherence rates. Among those in the at-home visual speed of processing training group, 28.3% completed less than one hour of training vs. 9.8%, 2.0%, and 3.2% in the on-site visual speed of processing without booster training, on-site visual speed of processing with booster training, and on-site attention control groups, respectively (p<0.001). A substantial amount of the non-adherence in the at-home training group was due to 32 participants' self-reported inability to successfully load the visual speed of processing software on their home PCs, a problem that would likely be resolved using the web-based delivery method that is now commercially available.

Table 1 contains the group means for the background characteristics of the 620 participants in the analytic sample at baseline, by training group and age band with p values obtained from analysis of variance or chi-squared tests as appropriate. The only significant difference (p<0.05) involved marital status for the middle aged (i.e., 50–64 year old) adults, with the attention control group being less likely to be married. This difference was also the only one observed when the age bands were combined (data not shown). Table 2 contains the group means (prior to Blom rank transformations) for the neuropsychological tests at baseline by training group and age band with p values obtained using Kruskal-Wallis tests. Significant differences (p<0.05) were observed only for the SDMT for the middle aged adults, with the attention control group being somewhat disadvantaged (i.e., fewer correct pairs). When the age bands were combined (data not shown), however, significant differences (p<0.05) were observed for the SDMT, Trails B, and Stroop Word outcomes, with the attention control group again being somewhat disadvantaged (i.e., fewer correct pairs, slower times, and fewer correct words).

Table 3 contains the results obtained from the linear mixed models of treatment group differences on the Blom rank transformed primary and secondary outcome measures. Cell entries are Cohen's d statistics (and associated Dunnett test p values) of the within-person changes in the neuropsychological outcomes from baseline to one year between each visual speed of processing training group vs. the attention control group. Cohen's d [42] statistics of 0.20 to 0.49 are considered small standardized effect sizes, Cohen's d statistics of 0.50 to 0.79 are considered medium standardized effect sizes, and Cohen's d statistics of 0.80 or greater are considered large standardized effect sizes. We note that there were no significant main effects for age bands, or for any interaction involving age bands. Thus, the standardized effect sizes shown were comparable for both age bands.

Compared to the attention control group, all three visual speed of processing training groups had significant small to medium effect size improvements (i.e., faster completion times) on the UFOV (primary outcome). The difference between the differences observed for the on-site visual speed of processing with and without booster training (at 11 months) groups vs. the on-site attention control group reflects the effect of booster training. The differences observed for the on-site and at-home visual speed of processing training groups reflect the comparability of the effect of these two training delivery modalities despite the lower adherence rates in the at-home group. To better characterize these effect sizes, Figure 3 plots the Blom rank transformed means for the UFOV by treatment group at baseline and at the one-year follow-up. As shown, the attention control group experienced a one-year age-related decline in UFOV scores. In contrast, the on-site (with no boosters) and the at-home training groups were equally protected from that one-year age-related decline, while the on-site with booster training group had improved UFOV scores.

All three visual speed of processing training groups also had significant small effect size improvements (i.e., faster completion times, more correct symbol to digit matches, or more correct words) on four of the secondary outcomes—the Trails A, Trails B, SDMT, and Stroop word tests. In contrast to the effects on the UFOV, there was no additional benefit observed for booster training on any of the secondary outcomes. Consistent with the effects on the UFOV, visual speed of processing training was comparably effective when it was received in an unstructured at-home setting as it was in the structured project laboratory setting. To better characterize these effects, treatment group means at baseline and the one-year follow-up were plotted (data not shown). In general, these means plots revealed age-related declines for the attention control group, but protection against age-related declines for the treatment groups on the Trails A, Trails B, SDMT, and Stroop word tests. Although there were no statistically significant visual speed of processing training effects on the other secondary outcomes (Stroop color and color-word, COWAT, and DVT time and errors tests), all observed differences reflected larger improvements (more correct responses, faster times, and fewer errors) vs. the attention control group that were also comparable across delivery modes.

To gauge the clinical relevance of the significant intervention effects shown in Table 3, we converted them into years of protection against age-related declines and/or years of improvement in cognitive performance. This was done by re-estimating the models using the original scalar values (i.e., non-Blom rank transformed scores), and then dividing those effects by annualized age declines determined from prior normative studies. As shown in Table 4, the significant intervention effects reflect protection against normal age-related declines and/or improvements of 3.0 to 4.1 years (depending on intervention group) on UFOV, 2.2 to 3.5 years on Trails A, 1.5 to 2.0 years on Trails B, 5.4 to 6.6 years on SDMT, and 2.3 to 2.7 years on Stroop Word.

Because of the significantly lower adherence rate in the at-home training group, the linear mixed model analyses were repeated after restricting the sample to those who completed at least one hour of training. With one exception those results were comparable to the analyses reported above. The exception involved the results for the primary outcome (UFOV). When non-adherent participants were excluded from the analyses the standardized effect sizes for visual speed of processing training were −0.278 for on-site training (p<0.05), −0.580 for on-site training with boosters at 11 months (p<0.001), and −0.524 for at-home training (p<0.001). Thus, when non-adherence was taken into consideration the observed effect sizes for the on-site with booster and at-home training groups were comparable.