Table 1 is organized in terms of the domains addressed and the types of treatments that were investigated and provides details about the methods and results. For the purpose of brevity, these studies are not described here in detail. Within the language domain, studies have typically provided a single domain intervention, such as naming, reading words or sentences and sentence production. As reviewed in Table 1, these studies include programs that target specific aspects of reading, naming or production.

To fully understand what the improvements observed across the various studies mean, it is important to provide an appropriate context. One way to look at improvement is to see whether the specific tasks delivered by the programs show improvement and to what extent performance on those tasks/sets of stimuli help improve performance on other tasks/sets (i.e., generalization). Another way to examine improvement is to see whether training on the tasks results in improvement on standardized tests. Different studies have looked at different measures on standardized tests, therefore, is it useful to classify the difference in improvement based on the type of measures that are utilized (i.e., impairment or functional/quality of life (QOL) measures). Figure 1 shows a schematic of how these outcomes relate to the targeted treatment. At the first level, treatment is expected to improve performance on the task trained, at the next level, this is expected to generalize to untrained items or other tasks. Next, the treatment might influence individuals' scores on impairment-based standardized measures, and followed by scores on functional and QOL measures. This progression illustrates if and how a particular treatment can have wide ranging impact on one's overall communication and QOL. Another factor worth considering is if the treatment has a long lasting impact on an individual's behavior (on either within task improvement or on standardized measures). This can be assessed in terms of whether studies conducted follow-up examinations (maintenance) and found the improvements to maintain over time on either within task improvements or on standardized measures (meaning that the standardized measures were administered multiple times and the gain found after treatment was sustained).

Of the 11 studies that focused treatment on only the naming domain (Bruce and Howard, 1987; Loverso et al., 1992; Aftonomos et al., 1997; Fink et al., 2002; Doesborgh et al., 2004; Raymer et al., 2006; Ramsberger and Marie, 2007; Fridriksson et al., 2009; Harnish et al., 2014; Kurland et al., 2014; Woolf et al., 2016), nine tested and found within task improvement, five of six studies that examined within task generalization found improvement, seven of eight studies that looked for gains on impairment-based measures found improvement, and the one study that examined functional/QOL measures did not find improvement. Additionally, four studies tested and found maintenance of within task improvement or maintenance of treatment-induced gains on standardized measures. Of the three studies that focused treatment on only the reading domain (Katz and Wertz, 1992, 1997; Cherney, 2010), none looked for within task improvement, one study examined and found within task generalization, all three tested and found gains on impairment-based measures, while none looked for functional/QOL measurement improvement. None of the studies tested maintenance of within task improvement or maintenance of treatment-induced gains on standardized measures.

Of the seven studies that focused treatment on only sentence processing or production (Crerar et al., 1996; Linebarger et al., 2007; Cherney and Halper, 2008; Manheim et al., 2009; Thompson et al., 2010; Cherney et al., 2014; Kalinyak-Fliszar et al., 2015), five studies tested and found within task improvement, three studies looked for and found within task generalization, four studies examined and found gains on impairment-based measures while two studies tested and found functional/QOL measurement improvement. Three studies looked for and found maintenance of within task improvement or maintenance of treatment-induced gains on standardized measures. Of, the two studies that focused treatment on the writing domain (Seron et al., 1980; Laganaro et al., 2006), one study examined and found within task improvement while the other tested and found within task generalization. Both studies looked for and found maintenance of that improvement.

Of the six studies that focused on multiple language domains (Mortley et al., 2004; Palmer et al., 2012; Corwin et al., 2014; Steele et al., 2014; Choi et al., 2016; Stark and Warburton, 2016), two studies examined and found within task improvement, one study tested and found within task generalization, three of four studies that looked for gains on impairment-based measures found improvement, while one study examined and found functional/QOL measurement improvement. Three studies tested and found maintenance of within task improvement or maintenance of treatment-induced gains on standardized measures. Of the four studies that focused on both language and cognitive domains (Wcislo et al., 2010; Hoover and Carney, 2014; Wenke et al., 2014; Des Roches et al., 2015), one study looked for and found within task improvement, none examined within task generalization, all four studies tested and found gains on impairment-based measures, while two studies found functional/QOL measurement improvement. None of these studies looked for maintenance of within task improvement or maintenance of treatment-induced gains on standardized measures.

To summarize, all the studies that have examined within task improvement have observed it, only one study did not find within task-generalization (Kurland et al., 2014). All but two studies (Steele et al., 2014; Woolf et al., 2016) found concurrent improvement on impairment-based measures. Fewer studies tested functional/QOL measures, and all but one study (Doesborgh et al., 2004) reported positive gains. Therefore, while there is a great deal of variability in the methods, design, and measures tested, the studies consistently demonstrated improvement and various degrees of generalization have been shown. Generalization can be considered a beneficial effect of treatment since it demonstrates efficiency for showing improvements beyond what was directly targeted. However, generalization can also be an issue when determining whether treatment was the reason for any improvement (Figure 1), which highlights the need for further research systematically examining each level of improvement across various rehabilitation programs.

One of the advantages of a computer-based rehabilitation is that it allows researchers to deliver rehabilitation the exact same way to all individuals, i.e., standardizing the delivery of rehabilitation. Such an approach could potentially deviate from the traditional ways of rehabilitation, where clinicians tailor their treatment for individuals. However, in traditional rehabilitation approaches, even though clinicians tailor the treatment, such rehabilitation is hard to replicate across patients. In this context, technology can provide the best of both worlds; technology can provide rehabilitation in a standardized manner and allows a clinician or researcher to tailor the tasks and items to an individual's deficits. Reviewing these studies allows us to examine to what extent studies deliver standardized treatment or tailor the treatment program for individual patients. For instance, ten of the studies (Bruce and Howard, 1987; Loverso et al., 1992; Crerar et al., 1996; Doesborgh et al., 2004; Laganaro et al., 2006; Fridriksson et al., 2009; Cherney, 2010; Thompson et al., 2010; Corwin et al., 2014; Kalinyak-Fliszar et al., 2015) required all participants to proceed through the same set of items and/or tasks. Therefore, rehabilitation in these studies was not tailored to individual patients.

Kurland et al. (2014) noted that their participants may have benefited from an increase in task difficulty after demonstrating improvement. This finding suggests that rehabilitation may not be best provided in a one-size-fits all manner, even if the item sets are tailored. Instead, rehabilitation should be dynamically adapted to the individual in terms of difficulty level, yet still be provided in a standardized way. Computer programs designed with algorithms allow for this manner of rehabilitation to be possible. Apart from the studies mentioned above, other studies in this review provided more individualized rehabilitation in terms of tailoring the item sets trained for each patient or by either allowing the clinician or researcher to tailor the tasks or the patients to choose which tasks to work on. For both standardized and tailored studies, the outcomes demonstrated generally positive results, however, it appears that individualizing treatment results in greater overall gains across the range of outcome measures, whereas studies that provided standardized treatment only measured and found gains in the within task measure, generalization, or impairment-based measures.

An important component of computer-based treatments is to allow for patients to practice the treatment outside of the clinic as part of their homework. This increases the amount of overall practice that patients can obtain and capitalizes on the ease of accessibility, which is another advantage of computer based-treatment. Several of the studies reviewed provided specific instructions for participants to practice treatment tasks at home. Seventeen studies (Aftonomos et al., 1997; Fink et al., 2002; Mortley et al., 2004; Linebarger et al., 2007; Ramsberger and Marie, 2007; Cherney and Halper, 2008; Fridriksson et al., 2009; Manheim et al., 2009; Palmer et al., 2012; Cherney et al., 2014; Hoover and Carney, 2014; Kurland et al., 2014; Steele et al., 2014; Des Roches et al., 2015; Choi et al., 2016; Stark and Warburton, 2016; Woolf et al., 2016) prescribed homework in addition to or instead of the treatment sessions. The remaining studies (Seron et al., 1980; Bruce and Howard, 1987; Katz and Wertz, 1992, 1997; Loverso et al., 1992; Crerar et al., 1996; Doesborgh et al., 2004; Laganaro et al., 2006; Raymer et al., 2006; Cherney, 2010; Thompson et al., 2010; Wcislo et al., 2010; Corwin et al., 2014; Harnish et al., 2014; Wenke et al., 2014; Kalinyak-Fliszar et al., 2015) only provided the computer-based treatment in the clinic when patients made their periodic visits.

To summarize, while studies that provide treatment only in the clinic show gains in outcomes, studies that provide homework are also quite effective. Only one study, Des Roches et al. (2015), compared patients who received homework to those that did not and found that the experimental patients (who received homework in addition to clinic time) improved on a greater number of tasks within the program and on a greater number of impairment-based measures. Future studies will need to examine these types of comparisons to ascertain whether providing homework treatment is as effective as traditional treatment and if there are any benefits to costs savings with home-based treatment.

Another potential advantage of computer-based rehabilitation is that, if indeed patients can easily access the prescribed rehabilitation wherever they have access to a computer, patients could potentially practice treatment more often. The question of whether increasing the intensity of rehabilitation is beneficial to outcomes has been a focus of traditional rehabilitation studies. While it is logical to assume that more intensive treatment results in greater outcomes and has been demonstrated in chronic (Bhogal et al., 2003a,b; Cherney et al., 2008) and in acute patients with aphasia (Godecke et al., 2014); other studies have questioned this premise (Bakheit et al., 2007; Dignam et al., 2015). For instance, a randomized controlled trial found that intensive treatment (up to 5 h/week) was no better than standard treatment (1–2 h/week) (Bakheit et al., 2007). Therefore, it is currently difficult to draw conclusions about what the optimal intensity and duration of rehabilitation should be for individual patients with aphasia.

Two studies (Raymer et al., 2006; Ramsberger and Marie, 2007) specifically varied the intensity of treatment. These studies found that all patients showed gains on within task performance and generalization to untrained items during both lower and higher intensity phases. Raymer et al. (2006) found greater gains in the higher intensity phase and, however, minimal improvement in impairment-based measures. Ramsberger and Marie (2007) found maintenance of the within task improvement. One study (Laganaro et al., 2006) also examined the effect of number of items treated as a level of intensity using a crossed design method, which found that, while the proportion of improvement was similar for both the smaller and larger (double the number of items) sets, there was a significantly greater improvement on the larger set, even though these items received less exposure in the treatment. This shows that exposure to a greater number of items provides proportionally greater amount of improvement than if exposed to a smaller number of items. Therefore, the higher number of repetitions in the smaller set did not result in greater improvement. Though this result does not contradict the effects of intensity in the previously mentioned studies, it might highlight the importance of greater intensity providing not only more repetitions for items but also exposure to a higher number of items.

Several of the studies reviewed prescribe a certain amount of home practice and in some of these studies, the homework was used to supplement treatment already given in the clinic. Twelve studies (Aftonomos et al., 1997; Mortley et al., 2004; Linebarger et al., 2007; Cherney and Halper, 2008; Manheim et al., 2009; Palmer et al., 2012; Kurland et al., 2014; Steele et al., 2014; Des Roches et al., 2015; Choi et al., 2016; Stark and Warburton, 2016; Woolf et al., 2016) prescribed homework to the patients but did not prescribe the amount/intensity of homework practice (instead suggesting a target time of practice). Of these studies, four used the homework to supplement clinic work (Aftonomos et al., 1997; Steele et al., 2014; Des Roches et al., 2015; Woolf et al., 2016), which potentially provided greater intensity of practice for the patients. The remaining eight studies used the homework as the only means of the therapy; some of the studies asked participants to practice for 20–30 min per day or on a specific number of days. If participants followed the prescribed amount of homework, they would not have received a greater intensity of therapy than if they came into the clinic, although one study (Choi et al., 2016) asked patients to practice at home for as often and as long as possible and found their participants received, on average, seven and a half hours of treatment per week. When examining treatment time across these twelve studies, home practice was not always documented but when it was, patients practiced for roughly four and a half hours on average [across the studies that report average usage (Aftonomos et al., 1997; Mortley et al., 2004; Linebarger et al., 2007; Manheim et al., 2009; Palmer et al., 2012; Kurland et al., 2014; Des Roches et al., 2015; Choi et al., 2016)]. These studies show that homework may be a way to increase the amount of practice that patients receive.

Additionally, in two studies, the control groups had inherently lesser frequency of therapy practice than the experimental groups. Des Roches et al. (2015) found that the experimental patients, who received homework in addition to clinic time, improved on a greater number of tasks within the program and on a greater number of impairment-based measures. Wenke et al. (2014) found that, while both the control group and the experimental groups showed gains on impairment-based measures, only the experimental groups showed gains on the functional/QOL measure. Taken together, these studies imply that greater intensity positively impacts treatment outcomes across the range of outcomes (Figure 1) that were examined.

Apart from the above mentioned studies, the other studies included in this review prescribed an amount of treatment time or criterion of performance to reach (while keeping intensity consistent). While the results from studies that prescribed homework and those that did not were both generally positive, one study (Choi et al., 2016) which allowed patients to decide the intensity of their homework practice found a positive relationship between the amount of practice with the program and scores on the K-WAB, where the participants who had a higher intensity of treatment showed greater improvement. Future studies should examine the relationship between amount of practice and treatment outcomes so that it can determined whether individually varied practice time can influence overall treatment outcomes.

Although only a few studies have tested the cost-effectiveness of technology-based rehabilitation compared to standard face-to-face rehabilitation in individuals with acquired brain injury/stroke, the studies that have examined this factor have found it to be a cost effective alternative. Palmer et al. (2012) calculated the cost of using the StepByStep program compared to the cost of usual care and estimated that the use of technology increased the cost over an individual's lifetime by a small amount. However, when considering the additional gains the experimental group made compared to the control group, the technology-based rehabilitation is considered to be quite cost-effective. Wenke et al. (2014) calculated the cost of standard service compared to three more intensive service models: computer treatment, group treatment, and treatment with a speech pathology therapy assistant (SPTA), and found the total costs were lowest for standard service and highest for SPTA treatment, with computer and group treatments being equal. However, once the cost of treatment per hour per client was calculated, the computer and group treatment models were the least costly, followed by standard service and then SPTA treatment. This is an area of further need of research if technological-based rehabilitation approaches have a long term future in the delivery of clinical services.

There is a significant financial relationship. Boston University owns a portion of stock equity in Constant Therapy, the software company that delivers the therapy. CD owns a portion of the stock equity that BU owns. SK is the co-founder and Scientific Advisor in Constant Therapy and owns stock equity in Constant Therapy. The results of the study are independent of the software platform and therefore there is no scientific overlap.